Mathematics Plugged

Wolfram|Alpha Pro

2012-02-09T00:00:00.000-08:00

is launched. A big step into possible individualizations.

What it will do.

It supports direct data, image, and file input - allowing to apply computational power to virtual any personal or public datasets and much more.

Wolfram Education Portal

2012-01-24T04:01:00.000-08:00

It is here. Wolfram Education Portal.
Nothing to add from my side, you know that I am excited about explorative, constructive learning and the fusion of mathematics and programming.

An interesting radical view: 100 reasons to run from lectures - identifying lectures as one of the 10 popular idols killing innovation.

Wittgenstein's Ladder

2012-01-03T07:47:00.000-08:00

I like Wittgenstein's Ladder in Tractatus Logico-Philosophicus: to go beyond you must throw away the ladder you climbed up (but you must have climbed up).

We are working on the All-New UnRisk.

Our UnRiskverse is fruit of an evolutionary approach.

First layer - grid supporting
For our numerically optimized pricing and calibration engines in C++ we have transferred the most advanced solvers from complex technical problem solving (chemical reactors, combustion engines, ..) to finance. Instead of the widely used tree-based or finite difference methods, we use Adaptive Integration, Finite Elements with Streamline Diffusion to solve the most sophisticated financial PDEs backwards in time for accuracy at speed, stability and robustness. Extended by (Quasi)Montecarlo with Longstaff Schwartz for models that can only be solved forwards. If possible FFT and wavelet techniques for in-milliseconds solving.
To avoid traps that are intrinsic in inverse problems we apply tricky regularization techniques in our calibration engine.
Adapting to the multi-core revolution we apply coarse grain parallelization.

All-New first layer - driving OpenCL over the grid
With the new computing muscles built by hybrid CPU/GPU architectures we can apply simpler general purpose algorithms.
As early adopters of this future technologies we are able to re-implement our code base optimized and platform-agnostisc - massively parallelized in OpenCL.
We can verify the new solvers by checking their I/O relations with those of the bank-proof solvers in C++.

Second layer - declarative programming
We integrated our number crunches into Mathematica and extended Mathematica into the universe of derivatives, portfolios, scenarios, models, methods, schedules, events, ... creating a domain specific language in the functional paradigm. From Mathematica, not common in financial circle at that time, we exploited two major principles: declarative programming and link technologies.
In not so rare cases, we develop UnRisk in UnRisk.

All-New second underlying layer - a unified description layer in C++
Having gained deep insight into the structure of a clean generic design by using Mathematica, we were able to decide to come closer to it by introducing new layer with a unified description of the financial objects and operations - in C++.

This layer will empower us to exploit the new support of C++ in Mathematica 8 in a generic sense and it will make the UnRisk Core even more open to multi-programming language integration.
For those who want to rely on bank-proof, blazingly fast pricing and calibration engines with a unified access from Mathematica in combination with other languages.

Us and creative copiers.

Fraunhofer IPMS -They Shape The Light

2011-12-15T02:02:00.000-08:00

The 2011 Future-Price of the German Federal President (Christian Wulff) went to Fraunhofer IPMS (Institute for Photonic Microsystems). With an Organic Electronics project.

This was celebrated yesterday in a show at a major German TV station (ZDF).

Congratulations!

Fraunhofer uses Mathematica in a general license since 1992. And IPMS does.

Cosma Shalizi: Why Economics Needs Data Mining

2011-12-02T04:26:00.001-08:00

From the INET interview series: 30 Ways to Be an Economist. Why Economics Needs DataMining.

Predictive modeling in economics is not so easy. We know the problem of overfitting and memorizing noise very well. And if the predicted values influence themselves you need clever algorithms and more frequent recalibration - computing implied features from reality and looking a little into the future.

This is one reasons why we work on an Econometrics package that combines Mathematica 8's excellent statistics and probability and our machine learning framework.

Computational Knowledge Provides More

2011-11-22T03:06:00.001-08:00

Yesterday, I completed reading Emanual Derman's new book Models.Behaving.Badly. A Book across philosophy, literature and high finance. He establishes the difference between model and theory to make clear that modeling financial markets cannot be scientific.

Emanual Derman is Head of Risk at Prisma Capital Partners and a professor at the Columbia University (financial engineering) and he was a leading Quant at Goldman Sachs.
I follow his Blog and I know that he is not only one of the most influencing quants (with Paul Wilmott he released the Financial Modelers Manifesto), but also with practical experience that includes where to find the best financial market data.

In his book he showed that the EMH (Efficient Market Hypothesis) and derived from it the CAPM (Capital Asset Pricing Model) are illuminating bat not reality - with an example of the comparison of the returns of Apple equities related to the S&P index over time.

A little surprise in the first second, that he did not take data from one of the leading market data providers, Blomberg, Reuters, .... to show the correlations required for explanation, but Wolfram|Alpha.
But after the input AAPL S&P 500 I saw it with my eyes. It is amazing.

One Day We Will

2011-11-11T01:19:00.001-08:00

Our mlf went and UnRisk goes cross platform. Windows, Linux, Mac OS X. Mathematica is cross platform, but our C++ kernel extensions were devoted to special target groups, like quant finance market participants, utilizing almost exclusively Wintel platforms.

But we go beyond. One day in the near future, we will have a kernel code that supports heterogeneous computing, like on hybrid CPUGPU platforms. One day we will have code that runs on a, say, DELL server with 128 CPUs and 4 NVIDIA Tesla 20 series of GPUS, my MacBook Pro wit 8 cores and an AMD GPU, my iPad or Android based tablets and other new computing muscles in the small and in the large.

Being early adopters of CUDA - Taming The Machine Infernal - we decided one year ago to overhaul the engines code base to make it compatible with heterogeneous computing platforms.
Core of the refactoring effort, real spade-work, is the radical redesign of all algorithms and their reimplementation in OpenCL. Yes this is a step back into programming in the small, but the platform independence pays back. The blazingly fast one-for-all code will run on current and future technologies. Transparent to developers, it will utilize the power of CPU cores combined with massive parallel architectures that define the parallelization granularity. OpenCL code is driven over the grid.

Developers still program in the domain-specific languages that are built in Mathematica's declarative language. Consequently, developers enjoy the double benefit of blazingly fast engines programmatically manipulated by a declarative language framework.
With Mathematica's OpenCLLink, we and the developers using our systems benefit from two great principles: first, messy plumbing of the utilization of hybrid platforms is automated (see the picture above). Second, what can be done with such a system can be done optimized without change - from a small device to a large HPC farm.

One day we will build all our software that way. And we offer training how-to already.

Machine learning framework goes OS X

2011-11-02T06:16:00.000-07:00

This is is the short announcement that our mlf is now available on Mac OS X. This extends the availability on Windows and Linux platforms.

Econometrics Learning

2011-10-17T06:20:00.000-07:00

Expectations play an important role in modern economic theories. 2011 Nobel price winners Sargent and Sims do research centered around the rational expectations framework.

The econometric learning approach, as I understand it, tries to model economic agents as forming expectations, on the basis of some learning methodology extracting and updating forecasting models. This means independent whether the expectations are rational.
As in machine learning in general such models might be extracted purely from data or combined with some mathematical forms that's parameters might be calibrated and re-calibrated towards concrete market behavior in an intelligent manner (in principle like outlined in the Blank Swan Of Metal Treatment, if I have this right)
But it might be that it is preferable that the agents start with little knowledge.

However, the methodological ingredients are in Mathematica 8 (comprehensive Statistics and Probability Superfunctions, fitness for the petabyte age) and mlf (creating understandable computational models from data in an multi-strategy and multi-method environment atop Mathematica).

This is the first time that I announce that we work on integrating both into a system that speaks the domain-specific language of econometrics and emphasize on the above mentioned learning aspects and hope to make systems more adaptive.

I find the challenges very well expressed in Learning in macroeconomics .. especially do we need to see market participants as players in the econometrics game, as we saw options trades as players of the game of Black Scholes options theory?

Idealism vs Realism In Programming

2011-10-13T02:06:00.000-07:00

We are hybrid programmers. We like to apply programming paradigms and consequently that fit to the problem type. Especially its decomposition paradigm. Data-oriented, function-oriented, object-oriented, logical, rule based, ....
But there is more: idealism vs realism?
Idealism strives for abstraction, expressiveness, productivity, portability, ....
Realism is driven by implementation, efficiency, performance, system programming.
The idealists love Mathematica, the realists C++?

There is a lot of pragmatism in Mathematica 8!

C/C++ interfaces:

CCompilerDriver - invoke C compilers from within Mathematica
SymbolicC - write and manipulate C syntax in Mathematica language, allows for programmable C code generation
LibraryLink - load C functions as native Matematica functions that become part of Mathematica kernel, share data between kernel and C functions. In an integrated workflow the CCompilerDriver package generates LibraryLink compatible shared libraries.

Execution envirinments:

Interpreter - includes everything but exclusively symbolic programming, arbitrary-preciasion calculation, heterogeneous expressions, ..
Compiler --> Wolfram Virtual Machine - optimizes for machine-precision calculation, homogeneous expressions, numerical functions, procedural programming, functionals programming
Compiler --> Native Code - transforming Mathematica function into a shared library using the above transformations

In certain applications speed ups can be 1:10 from Interpreter to Wolfram Virtual Machine and 1:10 from Wolfram Virtual Machine to Native Code (orthogonal to the support of grids and CUDA).

How to summarize and conclude?

Use Mathematica interpreter for productivity, compiler for performance. C language interface when necessary.

I have compiled this from a talk, our Senior Software Engineer, Sascha Kratky gave at the Austria Mathematica Conference recently.

We Speak at Wolfram Technology Conference 2011

2011-10-05T01:20:00.000-07:00

We not only attend the Wolfram Technology Conference 2011 but also speak. In "UnRisk 5 & Mathematica 8 - Blazingly fast an Insightful Analytics" we will highlight that Value at Risk (VaR) is not the end but the beginning of advance risk management processes. The VaR Universe of UnRisk 5 calculates a VaR cube for the deeper understanding of risk sources and impact of different risk factors. UnRisk's blazingly fast numerical schemes together with the computational power imposed by parallel architectures controlled by Mathematica 8 enable the calculation of such VaR cubes within minutes.

We will point out how the UnRisk-Mathematicas programming framework makes the development of customized risk management application quick and easy. This together with the advanced visualization techniques and the ability for rapid GUI development and dynamic graphics in Mathematica 8 allow for insightful risk analytics and reporting.

It is really amazing how our decision to use Mathematica as platform for UnRisk pays back. Even the most complex components can be tied together so easily. Driving CUDA over the Grid, transaction processing, dynamic visualization, task and work-flow automation, ...

We really look forward to joining Wolfram and other developers and users at the conference.

Constructive Math Learning Needs A Computer

2011-10-03T05:26:00.001-07:00

In Dance with Symbols I have summarized how I entered a business life driven by the constructive aspect of computer-based math and in It Is All Numbers about the indispensable numerical treatment solving complicated real life problems.
2009, in Courseware Builder I argued why I find Mathematica ideal for building courseware demanding that computer math is taught and not math. If our aim is to help students to become competent, autonomous acting experts in their field of practice we must give them the opportunity to behave as competent and autonomous as learner too.

Last week I enjoyed meeting Conrad Wolfram and heard about his great Computer-based Math initiative first-hand.
It is exciting to follow this project to build a completely new math curriculum with computer-based computation at its heart - enabling learners to concentrate on conceptually interesting topics and explore and construct knowledge.
Interesting to see that the constructive-learning theory (constructionism) goes back to the early 1980s.

Mathematics is essential, if you can do it by computer. Impossible to learn it without? Computer-based Math learning IS constructive learning.

Austria - Mathematica Conference 2011

2011-09-14T05:19:00.000-07:00

Things are moving so fast. We have now experienced working with Mathematica 8 since Nov-2011. At work we were inspired to thinking into new directions, like the combination of Mathematica 8's most comprehensive statistics and probability solvers and functions and our machine learning framework devotes to econometrics problems, or look deeper into econophysics studying pricing formation of financial instruments and the behavior of underlyers.
Financial risk management has already forced us to drive CUDA over the Grid in UnRisk.

On 6-Oct-11 we will present our and others experiences at the Austria - Mathematica Conference 2011 in Vienna.

The conference is packed with tips, tricks and techniques from developers and experts who have developed hundreds of thousands of lines of Mathematica code in hybrid environments with C++, Java, SQL, Web Services, ...

A Microscopy Application in Mathematica

2011-09-13T07:25:00.000-07:00

I enjoy reading the Wolfram Blog frequently.
The latest post Building a Microscopy Application in Mathematica, by Jon Mc Loone is really impressing (even not being a biologist). Simple and short code with great results.
It speaks on ist own.

There Are Many Traps In The Economic Landscape.

2011-08-29T06:50:00.000-07:00

I enjoyed meeting Michael Kelly, financial expert from Wolfram several times.
Recently on the Computational Finance Tour Europe 2011, where Michael, Andreas Binder from UnRisk and John Ashley from NVIDIA pointed out how important new technologies are that accelerate quantitative analytics and how excellent our techlogies are in interplay to do this.

It is my strong believe that our technology base will also positively influence high performance Econometrics working with immense collections of data and enormous computing muscles. With modeling and simulation, statistics, machine learning, dynamic visualization and more.

I read Michael's Blog thread Exploring America's Debt Problem and it convinced me again that Mathematica is the right technology platform to help controlling general economic systems by kind of macroscopes visualizing the dynamic of economic behavior and knowledge.

Mathematica Virtual Conference 2011

2011-08-05T04:34:00.000-07:00

The Wolfram Mathematica Virtual Conference has been announced recently. September 26, and September 27, 2011 to serve different time zones.
The virtual event will be packed with tips, tricks and techniques from Wolfram developers and experts. The agenda features 25 talks across 5 tracks.

Discover Mathematica
What's New in Mathematica 8
Applying Mathematica in Industry & Research
Applying Mathematica in Education
Developing on the Mathematica Platform

First we build the tools - Than thy build us, said Marshal McLuhan. Yes, this is true for the Mathematica Platform and its comprehensive dynamic interaction forms, including the direct way to create and play the Computational Document Format (CDF).

Connect Makers and Readers

2011-07-22T06:03:00.000-07:00

The Computational Document Format has been launched. To me it is the consequent continuation of multi-problem-facing in Mathematica: multi-strategy, multi-model, multi-method, multi-progammingstyles, multi-document. CDF are documents that come life with the power of computation.

See also Launching the CDF - Wolfram Blog, Conrad Wolfram. Nothing to be added from my side. Oh, yes, congratulations.

gridMathematica 8 - Driving CUDA Over the Grid

2011-07-19T06:33:00.000-07:00

Driving CUDA over the Grid posted Jon McLoone in Wolfram Blog recently. Mathematica's parallel computing design automates parallelization and anything that can be done in Mathematica can be done in parallel.

With the release of gridMathematica 8, which adds more than 500 new features of Mathematica 8 into the shared grid engine, one nice example brings together both ideas - that is driving CUDA hardware in parallel, over the grid.

We use this extensively in UnRisk multiplying speedups from massive fine grain parallelization (with a simple switch UseGPU -> True)and coarse grain parallelization (using ParallelMap, ParallelTable, .. commands). Together with an optimization of our proprietary algorithms we can achieve speedups in the many, many thousands.

UnRisk programming is high level programming in a finance specific language in Mathematica. Consequently it drives CUDA over the Grid.

Our algorithms, over 700.000 lines of C++ code, are numerically optimized for hybrid CPU-GPU systems, but programmers can manipulate them by an amazing few lines of UnRisk/Mathematica code including built-in parallelism. They are bank-proof in valuation and risk management of the most sophisticated deal types and complex portfolios. UnRisk atop Mathematica requires only 250.000 lines of code representing hundreds of instruments, dozens of models and methods and many nasty financial details and objects.

Driver Update

2011-07-08T04:26:00.000-07:00

(this is the title of a technology roundup in Wilmott Magazine, May 2011)
Each half year we make a roundup on how the drivers in the technology space will influence our future as makers and re-marketers of math-based software.

The need for speed, issues of model and method validation, data and computation management and the challenges of global accessibility have not changed. Challenges and opportunities of ease of programming and automation are not novel to us..
Summarizing it is the challenge of building adaptive frameworks that switch regimes, if requirements change. We do it ourselves and provide the foundations and know-how packages how to do it to our customers.

Modeling, model calibration and solving need to be adequate, accurate and robust. The ability to make them blazingly fast does not only save time but give us better ways to validate and switch models and work with adaptive precision. GPUs help us to achieve this and it seems NVIDIA Tesla architectures still drive the technology.
But the race for speed it also driven by coarse grain parallelism with grid and cloud computing - MS strives for simplifying parallelization up to the cloud hybrid with GPU support.
It is my strong believe that hybrid CPU-GPU desktops and servers will drive the business.

Combining analytics and data management still needs linking HPC computing and transaction processing in the SQL world

Domain specific programming, task and application building. This is THE new challenge.
All our packages are solutions and development systems in one. We provide domain specific languages and will make them available from standard programming languages.
Users manipulate our engines from certain user interfaces and/or programmatically.

Deployment. Although a transition into app-type front-ends is possible, we think of web deployment at the moment..

And still Mathematica is the only platform that has all ingredients we need. Mathematica 8 has not only convinced us with its great new capabilities in statistics and probability but also its software engineering tools enabling us to unleashing our innovations in environments users are experienced.

Consequently, accelerating our engines in C/C++ we will intensify the partnership with NVIDIA.
Grid and web computing, SQL links, domain specific languages, application building and link technologies will be atop the Mathematica 8 + platform. It is amazing, our stable partnership with Wolfram is in the 21st year now.

Our core competences: quant finance, econophysics, data mining and machine learning in solutions and symbolic computation, numerics, hybrid programming and HPC computing in techniques.

Wolfram Technology Conference 2011

2011-06-24T07:19:00.000-07:00

has been just been announced. It is again a must for us.
As developers of hybrid systems we look forward to absorbing news in technologies-in-HPC-environments and sharing experiences in CUDA programming (after 2 years of optimizing valuation and calibration engines for financial instruments combining new algorithms with massive parallelism on concrete NVIDIA architectures).
But also app and interface construction , web deployments and clearly computing with Wolfram|Alpha ...
And we look forward to meeting partners and friends.

Unleash Innovation

2011-06-17T05:08:00.000-07:00

Radical innovation sometimes needs to be rebuild from the bottom up. Businesses that redraw the landscape, creating needs we never knew we had are often only possible with technology jumps and not incremental innovation.
Whatever you did before, you might find better ways by utilizing new technologies - you put more intelligence in your foundations and relentlessly improve everything. And if you have made your products and processes innovative you might think of unleashing the innovation and position yourself in two-sided markets. You always were generous to customers, now you become generous to creative copiers.

At UnRisk we have transferred high-end numerical schemes from complex technical systems to finance. We introduces adaptive integration techniques and finite element extensions that were not common in financial circles. We act as lumberjacks, saying trees for solving financial PDEs are better as firewoods. But also naive finite difference applications ...
We integrated those numerically optimized C++ libraries into Mathematica and accelerated our development process drastically: The Short Story Of Being Lucky.

2 years ago we foresaw that utilizing the new computing muscles can reboot the innovative spiral. And it is not just reimplementation - we were engaged to find even better ways to solve the most complex tasks in computational finance. Four little NVIDIA Tesla C2070 cards (picture) provide us with the potential of a speed up of 400 performing valuation and calibration tasks compared to a contemporary CPU. But encouraged by this potential we introduced new methods that provide another speed up of 100 - that multiplies to a speed up of 40.000.

UnRisk 5 on Mathematica 8 both with CUDA support enable us to develop as declarative as before. To develop atop UnRisk/Mathematica will be as quick and allow for creating solutions in model validation, complex portfolio valuation, risk management and general financial decision support for uses we never have thought of.

To unleash innovation, we have decided to offer UnRisk-Q, now at version 5, our development environment supporting the latest computing technologies and algorithms to reduce time from hours to seconds, or enable billions instead of millions valuations in time.
To the big surprise an entry level turn key system UnRisk+Mathematica+a hybrid CPU/GPU machine will be only approx. EUR 13.000 - cluster power at the desk for a tenth of the cost.

UnRisk 5 on Mathematica 8 Utilizing NVIDIA Tesly 20 Series GPUs

2011-05-31T07:36:00.000-07:00

Yesterday, we have taken UnRisk 5 to financial institutions to valuate new deal types of inflation, mixed interest rates and FX, offer new stochastic equity models (variance and gamma and normal inverse Gaussian), make valuations blazingly fast by CUDA support and unleash the programming power behind the UnRisk suite.
UnRisk has been introduced 2001 - it is now its 18th release. This was only possible because of the careful design and Mathematica's high-level programming and link technologies.

Blazing Engines
The valuation and calibration engines are implemented by melting high-end numerical schemes and the next generation NVIDIA Tesla 20 series GPUs. With the stunning result that single valuations can be performed in microseconds enabling in-time calibration of complex models and scenario runs.

And Declarative Programming - For Creators
UnRisk-Q Version 5 is offered to quant developers as culmination of the co-evolutionary development of the bank-proof UnRisk PRICING ENGINE and FACTORY. Before its release UnRisk-Q was the hidden driver of the swift UnRisk growth - the major portion of UnRisk is programmed in UnRisk/Mathematica's task oriented, declarative programming language. Even the comprehensive VaR Universe producing cubes of VaRs across across risk factors and components has been developed with a fraction of the expected effort.

Making UnRisk-Q available to quant developers, we re-invented our business. They obtain blazing engines, programmatically manipulated in its task-oriented language, CUDA, grid- and web-enabled.

Software Prototyping is for Prototypers

2011-05-18T07:00:00.000-07:00

"Mathematica is a great system for software prototyping". Is it? Only? Or as well?
Many years ago, when the idea of software prototyping was born first we thought of throwaway prototypes. Later we were more specific and distinguished:

explorative prototypes - those to better understand user interaction patterns
experimental prototypes - those to clarify the decomposition type, software architecture and complexity by drilling down to critical functionality

both close-ending, thrown away. There are many other types and dimensions of prototypes with different notions.
With the availability of development workbenches and tools we went into

evolutionary prototyping - build the system by refining it and stretching and expanding its operational environment, this might be incremental or extreme ...

IMO, this is the past.
We now should think of evolutionary system development. Choose a universal working environment to create software systems in one flow by flatten the dimensions components, variants and revisions to revisions. Multi-strategy, multi-model, multi-datasource, multi-timeframe, cross-platform, multi-front-end, .....
Mathematica. It is a development platform integrating computation in any workflow supporting many technology platforms. If you view processes as computation you sit down, describe and get immediate results. This is the start point of an innovative spiral. It is more than refining, stretching and expanding.
Use the insight you get in step n to abstract, systematize, simplify or expand in step n+1.....

Clever in Any Climate

2011-05-11T06:45:00.000-07:00

Hot business - engines, usually providing a vast variety of high-end numerical schemes in a multi-strategy and multi-model environment as we find it in, say, process and risk control from chemical reactors, metallurgical plants to quantitative finance. Models usually are stochastic and partial differential equations solved by advanced finite element schemes to Montecarlo (better Quasi-Montecarlo) techniques. Usually the parameter identification (calibration) raises inverse-system problems that need special treatments, like regularization, ...

Blazing business - enabling the fast enough evaluation of such models (and this often means blazingly fast) one can forge the high-end numerics schemes and the next generation computing muscles as the NVIDIA Tesla 20 series of GPUs. The advances of such implementations can be measured in speed ups of thousands.

Cool business - to customize the analytics platform one wants to call the engines from a high-level declarative programming layer. For swifter time-to-insight you want to distribute tasks in a grid environment. Symbolic parallelization makes this easy. To increase transparency and reduce operational risk you need to apply the principle of consistent data and analytics management. In all this applications it is imperative to free info from silos, process analytics consistently and optimize systemwide transaction processing.

This requires link technologies and connectivities in your platform - With Mathematica we use the platform that empowers us to build large scale multi-model, multi-datasource and multi-timeframe systems.
Clever in any climate.

Now Ready For Registration: GPU Acceleration in Finance Events

2011-04-28T00:05:00.000-07:00

Wolfram, NVIDIA and UnRisk will present New Technologies For Accelerating Quantitative Analytics in London, Paris, Zurich, Frankfurt.
The UnRisk talk will focus on Efficient Valuation of Complex Derivatives on the GPU and give deep insight into the innovation I have outlined in Blazing Business.
The Registration Page.

I hope to meet you at one of the events.
You will benefit from

seeing how to manage complexity and reduce time-time-to-insight in quantitative finance drastically
live examples from the bank practice

Blazing Business With UnRisk 5 + Mathematica 8 + NVIDIA Tesla

2011-04-27T06:28:00.000-07:00

A derivatives professional values complex exotic options modeled by an advanced volatility model. The chosen model influences the theoretical price and it takes scenario runs to find the best one.
To calculate reliable prices each model needs to be calibrated to market data. She decided to use vanilla options on the underlying FTSE 100, Dax, Renault, Allianz, ... with volatility matrices for different strikes / expiries ... To do the calibration right and obtain parameters exhibiting a time stable and robust fit, clever optimization algorithms are a must ....
A single calibration task often more than 1 million single valuations. With traditional approaches this will most probably take 8 hours on a CPU. Considering that the process needs to be performed many times this is hopeless. Is it really?
At the MathFinance Conference in Frankfurt we have shown, how this can be done in 8 sec - see the Can You Feel The Heat presentation.

In about a month we will release UnRisk 5. It will have optimized CUDA support to solve the above task and alike blazingly fast. The engines are called from Mathematica 8. Mathematica 8 has CUDA Support which we instantiate with our proprietary algorithms.
In our comprehensive tests we utilized a 6 CPU Core machine with only one NVIDIA Tesla C2070 card. The derivatives expert should be already impressed with the time reduction from 8h to 8sec.

But, this is not the end - it is the beginning. If we multiply our grid computing speed-up with the new GPU acceleration, utilizing multiple NVIDIA Tesla cards, we can bring timing into fractions of seconds, for valuation and analytics tasks that have been assessed as undoable before.

This will create new perspectives for quant finance professionals and their customers and users.

Pre-announcement: Wolfram, NVIDIA and us will present the related technologies and the latest results in FREE Seminars on
6-Jun-11 in London
7-Jun-11 in Paris
14-Jun-11 in Zurich
15-Jun-11 in Frankfurt

Reserve the dates, I will update this info very soon.

Failure

2011-04-14T09:22:00.000-07:00

The HBR magazine April 2001 issue is about failure - understand, learn, recover. I agree that failure is a teacher, but is it the best teacher?
I also know that there are black swans and we cannot expect the unexpected.
And there are strategies to learn from failure and we should not be programmed thinking that failure is always bad.
But working at a,say, blast furnace, failures let you usually run.
As a coal-faced mathematician, I am socialized in industrial environments of flow processes and highly automated discrete manufacturing - hot and cool businesses, where failures might become disastrous.
Uncertain environments call for experimentation, but a stress test in a real plant is not so easy, you might need to learn from other secure processes in use and transform behavior into computational knowledge.

So, it is vital to understand the processes and its features to the very detail and especially what it means: running-outside-the-secure-properties. This requires modeling and inverse analysis, often in high-dimensional parameter spaces. We called this inline quality assurance and risk control.
Instead of measuring the product in production, control the process in relation to it. Unfortunately you do not always have a theory, so clever machine learning methods may help you to get insight and compute towards a high quality product with a secure process.
Mathematica 8, with its combination of modeling, statistical analysis and our machine learning extension will do a great job to avoid failure or detect them in its stage of coming into being. Our fields of experience are metallurgy, paper making, metal forming, assembling, ...
This might not be so easy in fields, where we have no theory (as in economy, IMO), but even there we shall strive for minimizing the hidden forces that shape our decisions?

Mad Science - The Cooking Lab

2011-04-05T06:50:00.000-07:00

In Wired march 2011, I read about Physics Meets Arts in the Cooking Lab. Nathan Myhrvold, Microsoft's former CTO takes on Modernist Cuisine. I am a lousy cook, but a quite talented eater. And I like to celebrate one or the other dinners at a famous chef's place. As Grant Achatz' Alinea in Chicago, or Heston Blumenthal's The Fat Duck, near London, both renown for driving the tools and techniques of molecular cuisine, mentioned in the article.
In If UnRisk was a Restaurant, I have outlined my view on the mathematical cuisine - that if you want to make the best dish you reduce processes by extending knowledge from the cooking knowledge base "correctly".
Myhrvold the scientist emphasizes on the processes. Chicken, salmon and beef are all coked sous vide with temperature probes inserted so Myhrvold could track how the heat moved through the food. He wrote a Mathematica program to model the heat transfer though various shapes and sizes of food.
His Team and its Tools - chef-reasearchers working in the most high tech kitchen ever created.

In short, cooks need to understand the physics of diffusion? However, ironically, avant-garde, science-driven cooking - is waning. Chefs are propagating breaks of tradition by serving, say, French cuisine with Asian influences. Or as my favorite chef Massimiliano Alajmo says, there-is-no-other-truth-but-the-one-of-the-ingredients (the axioms).
Maybe, I cannot suppress what Dan Ariely calls our-irrational-fear-of-the-unnatural.

New Course Assistants Released for Wolfram|Alpha

2011-04-01T06:38:00.000-07:00

31-Mar-11, Wolfram has released two additional apps on a growing list of Wolfram Course Assistant Apps.
I really like this approach for quick constructive learning and quick solutions.

Mathematica and Wolfram|Alpha - Fire Ignited In The Engine

2011-04-01T06:23:00.000-07:00

A car s a quite complex system. It can be viewed as a vehicle system, as a flexible multi-body dynamic system, and when it comes to the engine, a thermodynamic cycle system and a fluid-dynamics system. Modeling and simulation of the engine needs to solve flow problems with respect to geometric complexity and chemical and physical behavior. In the core it needs to deal with the combustion process and its heat and fire. And all its elements, components and systems are interacting. Control technologies range from mechanical, hydraulic to electronic.
As Stephen wolfram describes it in Launching a New Era in Large-Scale systems Modeling and making car models computable, needs large-scale systems modeling and simulation.
And yes, it is algorithmically demanding. And as in all systems of higher complexity you need to design and configure in the large as well as in the small. Processes and models need to be made computational and task-oriented descriptions shall manipulate them properly.
Mathematica is so fantastic at managing algorithmic complexity and MathModelica supports advanced multi-enginering modeling and simulation.
Having acquired MathCore Engineering, the integration of both technologies will become even tighter and possibly also empower Wolfram|Alpha to provide the widest access to engineering engines via gadgets, like the iPad. Congratulation to this acquisition!

Coming back to the combustion engines, the computational kernels need to be really mighty - it might be required that PDEs need to be solved in milliseconds and massive information needs to be transformed, aggregated and visualized, but the system design and programming can be driven by generic technologies in Mathematica.

In a comment to Stephen Wolfram's post, it is asked, whether this also would fit to stock market risk management. Yes the concept does. We have financial engineering, instrument structuring, portfolio construction, modeling and simulation across scenarios. The features are different, as well as the models and the solvers. And inverting plays a more important role. But everything else, has much in common.

We, at UnRisk, are proud hat we have seen the advantage of developing with Mathematica from the beginning. Where we are now, is amazing. Where we can go, even more.

Mathematica 8 - Fit for Econophysics

2011-03-29T03:08:00.000-07:00

In short, Econophysics wants to solve problems in economics applying methods from physics. One driving force behind econophysics was the emergent availability of a huge amount of data. Basic tools are probabilistic and statistical methods often taken from statistical physics. Moreover, IMO, attempts to use the theory of complex systems.
Working in quantitative finance, we know in depth that there are other tools from stochastic processes for pricing and risk analysis of financial instruments, analogies between financial theory and diffusion theory. The stochastic differential equations are often transformed into partial differential equations and depending on the deal types and contract features it is essential to have the right backwards or forward solvers and to treat the ill-posed inverse problem of calibration correctly - in UnRisk we worked for perfection of such solvers.
However, derivatives and risk analytics is characterized by the influence of a few factors (dominantly volatility of underlyings, correlations, ...), high dimensions come from the path dependence.
To study, say, market behavior or understanding systemic risk insights from the physical world can help, especially from systems in which networks of interacting units produce radical collective behavior.
Things like co-movements of, say, financial instruments are important - it is, for example, quite intuitive that in a panic co-movement increases - a stock market panic is like an avalanche where exogenous control becomes less important compared to streamline interactions within the network.

Mathematica 8 is fit for the petabyte age. It can manage massive data and has a vast variety of core algorithms related to probability and statistics and they become even stronger when intelligently combined with the many symbolic and numerical solvers. Methods that are only partially implemented or not covered can be easily extended or added. And we have the mlf that adds multi-strategy, multi-method machine learning.

We at uni software plus, have passionately worked in quantitative finance, data analysis and machine learning and we have extended our team with a theoretical physicist, teacher of an econophysics course, recently.
What have all our innovations, for over 20 years now, in common? Mathematica.
Econophysics is a field, where we get in quickly. It needs packaging an partnering.

Triumph of Hybrid Technologies - Mathematica 8 and C++ and CUDA

2011-03-16T10:05:00.000-07:00

We showed our latest achievements for quantitative finance markets at the Frankfurt MathFinance conference - See my Feel The Heat post in UnRisk Insight.
The problem in brief: to get insight into the model risk when valuing a complex financial instrument needs valuations across models. Models need to be calibrated to market data and each calibration might need one million valuations of simpler instruments to identify the parameters always checking the goodness of the fit of the model prices with the market prices.
Such single tasks can take hours on a traditional PC.
If those instruments are in portfolios and you need to test in scenarios you would need calculation time counted in weeks or months.
We have developed a clever combination of fast solvers in C++ and implemented parts on GPU systems (NVIDIA Tesla). Parallel execution on the GPU does not like if-then-else constructs, to make CUDA code efficient, it matters that GPU cores like it when they all do the same.

Mathematica 8's CUDA support is a perfect framework for this kind of efficiency tests - our C++ engines are seamlessly integrated into Mathematica and the CUDA support extends the link structure to CUDA implementations.
On top of this all we parallelized on the task level applying Mathematica's built-in parallel computing capabilities to the distribution of valuations across models and other scenarios.
This intelligent combination of coarse and massive fine grain parallelism is not only very flexible it also blazingly fast.

The reduction of 8 hours to 8 sec is possible on a 6 core PC with 1 Tesla C2070 card (448 GPU cores), but adding additional CPU and GPU cores we are fully scalable. We are now able to calculating results in a time that was assessed as impossible.
And we never leave our high level task-oriented language to put things together and just change and optimize on the lower levels.

Less Problems with Problem Solving

2011-03-03T08:48:00.000-08:00

Problem solving has been defined as higher-order cognitive process. Researchers have investigated problem solving separately in different knowledge domains.
With the focus on problem solving by computers, I characterize in analytic, knowledge-based and data-driven methods, of crucial importance in, say, engineering, applied to products and processes to take corrective actions to prevent failures or more general control risk.
They usually have trade offs in adequateness, accuracy, robustness, performance and economic aspects, like development efforts and resource requirements.
The most elegant models and solvers are often only valid in small domains and simple data-driven methods are resistant to generalizing.

Usually, one needs to find an intelligent combination of the three. The importance of knowledge-based and data-driven methods increases with the amount and complexity of data. R&D concentrates on uniquely combining statistical methods-machine learning-kernel-based techniques with cognitive approaches like fuzzy rule-based methods and we have successfully applied such combinations when controlling processes and their risk.

We have now analyzed and prototyped the new probability and statistical solvers in Mathematica 8, built in statistical distributions, distribution builders and parameter testers and I say, they are amazing.
They will open us new ways of solving even more complex problems in business and finance, process- and manufacturing engineering, energy supply, climate and weather, .. to public management, combining them with our machine learning framework for better analysis, prediction and control. Mlf is Mathematica 8 compatible already.

Ghosts in the Machines?

2011-02-22T07:02:00.000-08:00

Wilmott forums; a thread with that title started with the question of future programming languages in quant finance. And the possibility to implement "intelligent" solutions. Driven by dreams that the high frequency traders that know more much quicker, make their tamers rich?

When the Jeopardy match of the "Watson" computer against the human champions was started it was spiritedly discussed i the thread, whether the new computing muscles will influence algorithmic approaches.
(as you know, "Watson" pulverized the humans - knowledge mattered, as well decision and reaction).

In quant finance we have many models from the simple to the most sophisticated, related to deal types and contract features. Pricing with PDEs, we usually apply backwards-in-time solvers (partial differential equations are derived from the stochastic differential equations to manage this, knowing that instruments that's pricing is path dependent will create astronomic many paths, if going forward-in-time).

At the other hand, forward solvers are easier to implement, whilst pricing with PDEs one has to take care of pitfalls that come from the problem itself - usually Finite Difference or better Finite Element techniques are used. Even FE application needs some dirty tricks to make them accurate and robust in special cases. (however, avoid tree based methods for no-Mickey-Mouse cases).

But coming back to high-end computing. Will the new computing power change the way we build our computational engines? I think yes. In concrete the application of (Quasi)Montecarlo techniques will increase.
In general the computational engines will need less mathematical knowledge, but dancing with symbols will "move" into literal programming, task-oriented programming, domain specific languages and app building.
I do not know any other technology platform than Mathematica that fits so well for this two-sided technological revolution. Enormous computing muscles and language constructions for user programming and app building. The Wolfram|Alpha revolution the most impressive vector into this future.
UnRisk is uncompromisingly following this revolution.

1996 Checkmate on 10-Feb.

2011-02-11T04:16:00.000-08:00

From WIRED This Day in Tech: The first chess game between a human champion an a computer took place, with international Grandmaster Garry Kasparov losing to IBM's Deep Blue in Philadelphia.

We think that this was the first time when the symbol system human (the symbol system hypothesis states that intelligence operates on a system of symbols) lost to a kind of a brute force algorithm.

Wired: Aside from their stunt value, these man-vs.-computer matches have changed the way that chess is played, and not necessarily for the better. “We don’t work at chess anymore,” complained grandmaster Evgeny Bareev. “We just look at the stupid computer, we follow the latest games and find small improvements. We have lost depth.”

This reminds me on Marshall Mc Luhan: First We Build The Tools - Then They Build Us, he said.
Not bad, I say. Remove our irrational fear of the "unnatural".

Value-at-Risk Universe - Completely In Mathematica 8

2011-02-08T06:38:00.000-08:00

We are in the finish line of an unprecedented comprehensive VaR module that we have developed on top of UnRisk-Q. I wrote about its motivation, the state-of-the-art and the achievements in VaR of the Jungle.
On top of UnRisk-Q means that we used financial objects that are represented in the Mathematica language and programmed all higher level constructs representing the VaR universe in Mathematica.

Through different layers it uses numerical schemes optimized in C++. This empowers us in combination with Mathematica's symbolic parallelization techniques to calculate VaR cubes of large portfolios across hundreds of scenarios in a few minutes.
But the development took a very small group of Mathematica programmers only about 4 months. The innovative spiral is accelerating with each development.
Quant developers and risk professionals might only need a few lines of high-level code to develop a back test, stress test or other risk test environment for their portfolio- up to firm-wide risk assessment.

The Beauty Of Software

2011-02-02T06:45:00.000-08:00

I receive international science grid this weak . With a link to the beauty of software today.
The visualization of software version control is not new, but has been further developed nicely.

Remove The Irrational Fear Of New Architectures

2011-02-02T06:03:00.000-08:00

Our senior software engineer's responsibility is to design our products for technology platforms with future.

2010 saw the beginning of the end of the Wintel area. With Microsoft adding support for ARM processors in Windows 8 and many hardware manufacturers moving away from Windows and choosing Android as an operating system platform for mobile computing devices the computing platform landscape will change considerably in the coming decade.

One of our major tasks in 2011 with be the overhauling of the UnRisk engines code base to make it compatible with target operating systems other than Windows and to support modern processor architectures like ARM or Atom.

The UnRisk engine code base consists now of around 700 000 lines of C++ code and an additional 200 000 lines of Mathematica code.

The first quick outcome of this refactoring effort will be a Linux version of the UnRisk engines which will work seamlessly with all Linux platforms supported by Mathematica 8.

To make the box fast as lightning, we support hybrid CPU-GPU architectures from NVIDIA to do things that cannot be done without.

And clearly, we will continue our spade-work providing tools for integrating UnRisk and web services (webUnRisk) and mobile device support being ready for the SaaS/web revolution.

90 Years Ago - A Play About Robots Premieres In Prague.

2011-01-25T08:33:00.000-08:00

Wired blog This Day In Tech again. Jan-25-1921: Robots First Czech In.
Karel Capeks play marks the first use of the word robot (forced labor). The robot in the play are not mechanical men made metal. Instead they are molded out of chemical batter and they look exactly like human.
This reminds me on on a seminar work in one of the famous robotics labs (forgot which, sorry) in the 1980ies. Students were ask to design a thing that washes dishes automatically. All kind of mechanisms with redundant axes and complex kinematics were designed ....
But we had dishwashers already at that time.
So, when I was responsible for factory automation, I always tried to "preach": simplify-first-then-automate.
But pick-and-place tasks often forced us to apply robots with redundant axes and the inverse kinematics problem (given a set of paths, searched the kinematic system .. ) became quite challenging.
With the computers of limited performance a symbolic computation system would have been great to automate the modeling process itself.
Today, automation, control and control programming is built much more bottom-up (see also The AI Revolution Is On ).

The Race Of Speed

2011-01-25T04:48:00.000-08:00

Struggling to keep up with the pace of innovation a technological arms race has become dominant in various industry sectors.
Take high-frequency stock trading where algorithms zip in and out of markets often changing orders within seconds. Part of all stock trading takes place in a warehouse, filled with long avenues of computer servers, close to New Jersey Turnpike. But high-frequency traders do not only use algorithms, they also speed-read news to change strategies within seconds. Like .com companies were the reason for bandwidth, high-frequency traders are the reason for massive computing infrastructure.
At the other hand new processors are under way based on hybrid CPU-GPU architectures, like Fermi from NVIDIA. Those architectures offer supercomputer power under the desk and later on tablet computers.
If we remove our irrational fear of new architectures and use such new computing muscles what does that mean to scientific and technical computing?
First, we can do things, we could not do before - like real-time analytics by frequent model calibration and re-calibration.
Second, we might save time and use more brute-force approaches instead of developing clever algorithms - like Montecarlo techniques to solve all type of PDEs, PIDEs, ... or work in high-dimensional parameter spaces instead of the real world.
Third, to exploiting a deluge of data we might use other feature extraction techniques than mathematical modeling.
But it is my strong believe that as massive computing muscles emerge they will also open doors for new human-machine intelligence business.
But we also need to remove our irrational fear of new hybrid software platforms and programming techniques.
Sometimes we might say: mathematical code is dead - long live mathematical thinking and free-form programming?

Fruit Of The Loom

2011-01-19T03:26:00.000-08:00

The cloths brand position themselves as vertically integrated manufacturer - they manufacture their own yarn, knit the cloth, cut the fabric, sew the garments and package the products themselves.
I do not know enough of this business to understand whether they apply made-to-measure approaches - defining products that are made from a standard-sized base pattern but customized for individual requirements. However.

The Bottom-Up approach of Artificial Intelligence.
Intelligent behavior is reconstructed starting with the most simple functions/programs, interweaved in superior layers causing the emergence of complex structures.
This is in contrast to the approaches derived from the symbol system hypothesis that intelligence operate on a system of symbols by reasoning or inference engines.

Mathematical Problem Solving.
Symbolic computation is promising closed form solutions (of good nature for exact results), but the "worlds" its models describe are often too small. Numerical approaches look much more bottom-up by decomposition of domains and special mathematical functions into simple mathematical behavior. It usually covers much more problem domains and classes.
Ironically, in an inverse problem view you look for models that predict behavior you have observed or want to control. Symbols will help you to understand, calibrate, ..

I have mentioned that we face such problem in advanced quantitative finance - UnRisk. Mathematica is indispensable for our ambitious approaches. Not only that it covers all of the above paradigms in a uniform way and high-level language representation, its unprecedented software development support, significantly enhanced in Mathematica 8, allows UnRisk to do things that cannot be done without. And, when we take UnRisk/Mathematica to quant finance professionals in financial institutions they do things the Mathematica way, which Mathematica could not do ....
I call this co-evolution.

What Is It What Mathematica Can't Do?

2011-01-13T01:01:00.000-08:00

The universality of Mathematica does not mean that it solves each special problem instantly. If you have a system of coupled tricky reaction-convection-diffusion PDEs you might need a tricky solver that enhances, say, finite element techniques. We have such animals in advanced quant finance, processes in chemical reactors have, ....
But,
Mathematica does so many things FOR YOU that I would need far too many pages to describe. You sit down and compute symbolically and numerically, visualize ....
Mathematica offer to do infinite many things WITH YOU by enabling you to make programs out of your computation in an unprecedented intuitive declarative manner and a seamless workflow ...
and
You can do things FOR MATHEMATICA (close to perfection in version 8) by integrating your own specialized solvers into it. Doing this you extend the Mathematica kernel and your special objects speak Mathematica. Making it a package and offering it to the market you extend Mathematica into a special domain universe.
Like UnRisk extends Mathematica into the universe of advanced financial derivatives and risk analytics - the UnRiskverse - instantly used by traders, treasurers, asset and investment managers, fund managers, risk controllers and risk managers and programmatically extended by quant developers in banks, capital management firms, funds and insurance world-wide.

That can trigger innovative spirals you cannot drive in any other environment (to my knowledge).

What Is It What Only Mathematica Can Do?

2011-01-12T05:53:00.000-08:00

This question arose when I read in WIRED's Blog This Day In Tech 12-Jan-92 or 97: HAL of a Computer, 1992: HAL 9000, the master computer aboard the Discovery spaceship in the novel and film '2001: A Space Odyssey', becomes operational. He will inspire millions of dreams — and some nightmares — of artificial intelligence.
An artificial intelligence that could think, talk, see, feel and occasionally go berserk? HAL - a fictional computer that is more advanced than any computer today?
From the The AI Revolution Is On one could summarize: systems in the forefront of artificial intelligence shall apply machine learning and massive data management instead of emulation human intelligence.
And beyond new ways to program and compute are required.
What is It .. ?
IMO,

Natural language input and in general the possibility to define any syntax and input form that generates programs that trigger anything from simple to complex computation. The strength of linguistic capabilities has been shown by, say, fuzzy logic, but its representation is restricted to rule-bases
More general the tight integration with Wolfram|Alpha
The capabilities to dancing with symbols and numbers in a way that allow for asymptotic math approaches supported by automated algorithm selection and adaptive precision control (symbolic parallelism, for example, saved us so much development and optimization efforts)
Pattern matching, let functions be applied to practically anything. It is indispensable if you want a program layer that describes in the language of mathematics

I might have missed other capabilities, but I think, this is a lot ....?

The AI Revolution Is On

2011-01-04T04:43:00.000-08:00

Title of a WIRED essay by Steven Levy. Stephen Levy's Artificial Life is one of my favorite books on science and technology. BTW, much of the book deals with cellular automata and Stephen Wolfram's revolutionary work - peaking a little later in NKS.
Artificial Life was released in the early 1990ies.
It also mentions systems with many objects that have little intelligence themselves but intelligent behavior in interplay (kind of ants-on-silicon).

In the essay Levy entries with the description of warehouses that are managed by robots automatically.
Like Kiva Systems . Levy: The Kiva bots may not seem very smart. They don’t possess anything like human intelligence and certainly couldn’t pass a Turing test. But they represent a new forefront in the field of artificial intelligence.
Instead of emulating human intelligence and re-create the human brain, systems use machine learning and massive data.

But, IMO, it is not only dancing with machines. In Dance With Symbols and It is All Numbers I have discussed two concepts that look controversial at first place, but combined to Asymptotic Mathematics it might allow for the adaptiveness required for intelligent systems.

I think of a system that does continuously compute where it thinks that it found information for computational knowledge. It produces massive useful results. And a machine learning system extracts and calibrates knowledge-for-decisions of robots, agents, .. that re-calibrate it to the absolute of reals life behaviour.

Bursting with technologies, described Stephen Wolfram Wolfram Research at the Wolfram Technology Conference 2010 .
And in his Blog post Future Directions Of Wolfram Technologies : you will soon “see a lot of different directions emerge” based on the technology and technology platforms that Mathematica and Wolfram|Alpha provide.

In our UnRisk business we work towards a risk management FACTORY that continuously computes tons of prices, theoretical returns and risk spectra of portfolios-across scenarios, risk factors, .. all based on calibrated models and re-calibrated to market data of liquid instruments. Only information that is relevant to the trader or risk manager will be extracted and prepared adequately for decision making.
If the trader and risk managers become computer programs themselves - nothing changes.

Fortunately, we have selected the right technology platforms (Bursting with technologies ...).

The Blank Swan Of Metal Treatment

2010-12-22T03:43:00.000-08:00

Finance. This year Elie Ayache released his controversially discussed book The Blank Swan, that presents a revolutionary way of treating financial derivatives valuation and technology. Its title does relate to NN Taleb's Black-Swan-Events (unexpected events that bring a deal in a new regime of anomalies and risk), but with a reverse-probabilistic perspective. In fact, the thought behind Ayache's book is deeply challenging. In short, a financial derivative (Ayache calls it a contingent claim) materializes as price in a market only, if a trader trades it. Everything else is inverting.

Is the use of a valuation tool, like our UnRisk, merely a means to inert ourselves into the market, recalibrate a model against the market and then repeat the procedure? If it comes to options, I let Ayache speak ... It is only because the dynamic trader has to be hedging continuously his position in the option and therefore is following the option continuously; it is only because of this that he is basically entitled to compute something we call implied volatility - and implied volatility is the simplest instance of recalibration ... recalibrating the Black Scholes model to the option market price ....

Traders do not price because of models, but get insight into the context that has been written into the BLANK contingency - that led to the price. A Black Swan then is only a context switching event of the same thing and a context switching regime is "just" another recalibration approach.

In the picture above, we have a Heston model, an extension of the BS model with stochastic volatility, in PDE form. To calibrate and frequently recalibrate such a model is really challenging.
You need to do things in seconds that might take hours with traditional environments and implementations.
We at UnRisk, do this with ultrafast numerical schemes, principal component application, grid-computing, CUDA support, ... all managed and linked by Mathematica, now on version 8.

How could we achieve this? We met Blank Swans before we entered into quant finance.
Let us take a numerically controlled metal sheet forming machine. It "trades" a shape and the return pays in accuracy of the shape. To implement the control, you need to understand the framework of the elastoplasticity theory and the complexity and limits of its mathematics derived from the mathematics of continuum mechanics. Deformation is decomposed into elastic and plastic parts and for simplicity decompositions shall determine stress and kinematical quantities.

The resulting PDE system can be solved by advanced numerical schemes.
They "only" need to be calibrated related to the physical properties of the material that are dependent on recipes, and properties that are result of the process? And work as predictive models for final shapes? NO.

If you buy a metal sheet you order it by standardized "names". And this is where the headache begins.
Standardized mass-steel specifications of one material instance allow ranges of physical properties of up to 15%. But each single sheet has its own metallurgical fingerprints.
It is impossible, to measure the properties before bending, because you needed to destroy part of the sheet.
So, you need to recalibrate your models during the forming process ("continuously"), say, by observing force-shape trajectories over the process. For explanation of the quantified elasto-plastic behavior in this step. The closer you come to the final shape the more your system knows about the concrete material and the better it can explain and predict a little into the future.

Calibration here is often result of a machine learning application with our mlf and recalibration uses the extracted knowledge (instead of volatility, we compute implied material parameters).

This is a paradigm for all kind of Blank Swan processes, where modeling, ironically, is not for prediction, but explanation and the computation of implied features.

Mathematica is the perfect environment for this. It let you create understandable and computational models, it supports cross-model testing and integrates your ultrafast solvers seamlessly.

How Much Beer

2010-12-10T06:44:00.000-08:00

I remembered Dec-09, I found an article in Wired, Geologists Love Beer and I asked Thomas Natschlaeger whether we have something about Beer. He quickly wrote this notebook: how to build a model in machine learning framework to predict the Australian beer production.

Part 1

Part 2

Part 3

UnRisk on Mathematica 8 Crunches Numbers on Little Hardware

2010-12-02T08:50:00.000-08:00

In Mathematica 8 and UnRisk, I have highlighted the benefits from utilizing new features in Mathematica 8. Being curious, I have set up a quick test on a one-core Windows laptop with an nvidia graphic-card.

The problem: to valuate an exotic option under a stochastic volatility model (Heston in our case), you need to calibrate the model towards market prices of, in our case, 256 vanilla options. To make this quick for test purposes we used the ultra-fast Fourier-Cosinus method for valuation.

I tested UnRisk in 4 variants, and recorded absolute calculation time in seconds

a) UnRisk C++ engines integrated with MathLink - 9.1

b) UnRisk C++ engines integrated with MathLink + GPU utilization - 0.3

c) UnRisk C++ engines integrated as DLL - 9.0

d) UnRisk C++ engines integrated as DLL + GPU utilization - 0.2

Click the picture, to find the quickly drafted code of the calculation part in the Mathematica notebook.

In this synthetic example I need little data but an extensive number of valuations, consequently the speed-up of the DLL integration is negligible. In real cases this will pay much more.

But the speed-up from utilizing the power of a little, low-cost HW extension is already enormous.

And it is so convenient having Mathematica as a universal interface to GPU programming.

Programming With Natural Language

2010-11-19T07:15:00.000-08:00

In Domain Specific Languages ... I mentioned that task-oriented languages have been introduced quite early in factory automation. I have been responsible for factory automation software in Austria's largest industries in the 1980ies. We have developed off-line programming systems for machine tools, robots, automated loading, ... and flexible manufacturing systems that integrated them all.
We offered "Grasp Part", "Mill Pocket", .. constructs -that were interpreted- instead of "MoveTo(x,y,z,u,v,w)".
To transform higher-level languages into a representation for actions, we had algorithms for inverse kinematics, inverse dynamics, path builders and so on.

Today I read Stephen Wolfram's post Programming with Natural Language is Actually Going to Work. It is amazing. This goes much beyond my imagination.

In the 1980ies we did not have the technologies, to go further than our task-oriented, but still strictly formal, syntax. But, I need to confess, that I also did not have the knowledge on the principles of languages, beyond formal syntax and compiler techniques.

I am really deeply impressed about Wolfram's achievements. Congratulations!!!

It seems Mathematica in co-evolution with Wofram|Alpha will set a pace that will be difficult to be followed by competitors.
We will take the advantage of making our solutions more intelligent and intuitive. Happy to having selected Mathematica 20 years ago.

Wolfram Takes Mathematica 8 To All Of Us Now

2010-11-15T10:23:00.001-08:00

The new release "materialized" on the Wolfram Page . Just now.
It has so many new features and some of them, probably, have never been done before. Free-form linguistic input fruit of the Wolfram|Alpha integration as well as the combination of knowledge and computation with the provision of over 10 trillions of curated data .......
It is amazing. Among so many great new features one thing might be even be a bit underestimated? Performance.

For us developers and solution integrators Mathematica 8 brings a full basket of software engineering gifts. In short, CUDA support, direct integration of DLLs and fast and auto-parallelized compiled-into-C-code functions.
It will not happen in a day or two, but we really will "fly" to dance with those features. For UnRisk and the machine learning framework .

It will allow us for building UnRisk power stations for quantitative analysts that enable them for calibration and continuous recalibration of complex models for sophisticated deal types and portfolio across scenario testing.
To do this in-time calibration time need to be reduced from 8h to 10 seconds - in a box that is still a "PC".

Boosting calculations without leaving our strategic position of driving generic technologies is possible. With Mathematica and our optimized numerical schemes, co-created.

The Next Scientific Revolution?

2010-11-08T03:08:00.000-08:00

A feature in Harvard Business Review Magazine, Nov-10, by Tony Hey, Microsoft Research.
Because of our ability to collect and analyze vast variety of data, scientists now have the potential to solve some of the world's biggest problems. But until very recently, we didn't know how to combine the right data sets and see crucial patterns, the article describes the idea in brief.
It was probably the first time, since I have subscribed HBR, that I read "Machine Learning" in a paragraph header.
Machine learning is an "old" methodology, but its application to real live problems took a while, because they need massive data and enormous computational power.
Why? Because most of the systems that can attack problems in life sciences, energy management, climate change, process industry, .... need multi-strategy and multi-method approaches.
I share the enthusiasm about the possibilities in the petabyte- and petaflop age. But there are some barriers and limitations. In short, we want to extract knowledge and computational models from data. For

Analysis
Prediction
Control

In analysis, we want to get insight. Usually this is not so difficult.

In prediction, we want to create models that are able to predict future behavior. In machine learning generalization is difficult.

To do predictive modeling adequately, we usually need to introduce meta-learning algorithms and combine models. Mathematica's programming environment is ideal for this. It also provides built-in parallelism..

We exploit this technologies in our mlf .

In control we want to manage parameters that lead to desired goals (like quality features). To do this we need to solve inverse problems that are ill-posed by nature. The machine learning part is often focussed on the support of parameter identification in model calibration. Machine learning interact with theoretical models. To do this, you might need symbolic and numerical computation and machine learning in the framework of Mathematica.

The metal forming example

Take free bending. You need to understand the framework of the elastoplasticity theory and the complexity and limits of its mathematics derived from the mathematics of continuum mechanics. Deformation is decomposed into elastic and plastic parts and for simplicity decompositions shall determine stress and kinematical quantities.
However, resulting PDEs can be solved by finite element schemes. They "only" need to be calibrated related to material properties.

If you buy a metal sheet you order it by standardized "names". And this is where the headache begins.
Standardized mass-steel qualities allow ranges of physical properties of up to 15% in one product. But each single sheet has its own metallurgical fingerprints. Measuring the properties before bending would destroy part of the sheet.
So, you need to recalibrate your models during the forming process ("continuously"). The closer you come to the final shape the more your system knows about the concrete material and the better it can control ....
Continuous measurements of the shape? The material does not stay in the form you brought it by the machine, it uses its elasticity memory and reshapes.

We have used mlf on Mathematica to solve such problems for concrete machine controls.

The Beauty of Innovation

2010-11-07T05:01:00.000-08:00

We, at UnRisk, spoke at Wolfram's Analytics & Risk Technology Conference. In our talks we presented results that were only possible because of 2 innovations

the transformation of advanced numerical schemes from complex technical system solvers to finance
their seamless integration into Mathematica

Does beauty exist in an novel and elegant approach? A system that works intuitively, accurate, robust and efficient? Yes? Then our innovation is beautiful.

When I think of the other practical presentations from Royal Bank of Scotland (Derek Yates, Corporate Debt Risk Management) and King's College (William Shaw, Optimization of General Risk Measures) , we all seemed to have in common: creating beauty is not difficult, provided your foundations (let me call it kernel systems) are adequate. Mathematica presents, manipulates them and ties them together in an unprecedented elegant and efficient way. Beautiful.

No longer the world should expect the representation of complex systems in complex technologies. The beauty of innovation will matter.

Not In The Toothbrush Business

2010-10-19T09:35:00.000-07:00

15 months ago, I wrote about this in UnRisk Insight - Not In The ... .
I talked about the temptation to offer an option calculator for everybody. But, we knew, that the Black Scholes formula needs to be generalized and its extensions with stochastic volatility and jumps require sophisticated numerical schemes and calibration techniques. Now, the quant finance experts come to the conclusion that calibration (re-calibration) is the key to keep the dynamic replication of an option right. A few days ago, Emanuel Derman, one of the most influencing quants, wrote about this in Calibration .
Since Elie Ayache wrote his "The Blank Swan", calibration might become even more important - see Market Of The Future, Or Market With Future?
The UnRisk makers are good in calibration, because on of their scientific background in "Inverse Problems".

At the other hand we have the Mathematica front-end, where quants and risk professionals can run scenarios across models and methods (we call this orthogonal organization). And if prices and risk spectra differ too much, something might be wrong with the deal type? This has also the positive effect of explorative learning that keeps the product use learning cycle unparalleled short and often self-paced. No need to "delight" them with long guided training sessions.

We Did It Already. The UnRiskverse.

2010-10-18T06:13:00.000-07:00

I have outlined a process for building an application universe . It is fruit of challenging research and development. When building the UnRiskverse, the UnRisk makers , followed exactly the 7 Acts.
I develops quick, because UnRisk is often made in UnRisk and Mathematica.
A customer once pointedly asked: how can you deliver a new deal type with full valuation capabilities, before we get an inquiry answered by somebody else?

It is a dynamic process.
This is what we have achieved at the moment. The multi-language UnRiskverse that is not a Babylonian language confusion.
In lines of code:

Act 1- C++: 650.000 representing the most sophisticated numerical schemes (that are not common in financial circles), like advanced assymptotic-math techniques, FE with streamline diffusion, regularization to do the inverse problem of calibration right, code that has been optimized for CUDA architectures, ..
Act 2 - Mathematica: 200.000 for the treatment of deal types, portfolios, scenarios, VaR; valuation, analytics, visualization ..
Act 3 - gridMathematica: a few lines of the built-in parallelization techniques
Act 4 - SQL: 170.000 in combination with Mathematica's data base tool kit and Java
Act 5 and 6 - 400.000 lines in web-supporting languages (Java, Java Server pages, XML, ....) web-enabling Mathematica

Isn't it horrible to test in such a hybrid environment?

No, we are producing Mathematica Notebook Logs from any action automatically and re-feed them in test and support cases.

What is this architecture made for? Number crunch in London, valuate in Singapore, back-test in NY. Access UnRisk from your desktop, laptop, iPad, ... UnRisk follows its user. Develop your own variants and systems with a bank-proof foundation and an intensively tested architecture.

Tired Of Traditional Technical SW Development? Try This Instead.

2010-10-15T05:38:00.000-07:00

In the mid 90ies of the previous century time became ripe for developing PC-based software even for challenging quantitative problems in all kind of technical fields. Many efforts were put to become virtuosos in C+, C#, ... to transforming complex models and algorithms into LAN-enabled systems.
But there are much better ways now to build your special application universes.

Act 1 - Set up the foundation - apply the methods and programming languages that fit for your purpose. If you need sophisticated numerical schemes you might choose C++, C#, ...
Act 2 - Integrate your foundation into Mathematica - in many technical fields your objects can be described in the language of Mathematics and manipulated and operated mathematically. Your foundation will extend Mathematica into the universe of your domain.
Act 3 - Set a distributed computing environment to work. Facing the multi-core and multi-thread revolution (like CUDA ) you multiply performance by a clever mix of symbolic coarse- and fine grain parallelization. gridMathematica supports scalable infrastructure.
Act 4 - Set up application data base management - you often want to have technical data encapsulated.
Act 5 - Build a web connection layer - webMathematica and web a services will help you to wrap your Mathematica and make it a Servelet web application. Build your adequate web interaction pattern.
Act 6 - Tie everything together. Your server application (in the cloud?) - data base (local?) - web front-ends (including special "iPad", "iPhone", ... front-ends?).
Act 7 - Make it scalable by license management and business modeling.

It might become your short story of being lucky.

The Internet Of .... II?

2010-10-12T08:00:00.000-07:00

We want to solve more or less complex problems and provide solutions in many fields from engineering to science. We do not only compute, but model, simulate, analyze, visualize, identify parameters, recognize features, report results ...
If we do algorithmic mathematics, we need to develop and we want users, consequently, we deploy.
We do not want to integrate versions of computers, operating systems, tools, standard applications, ...
We want to build generic technologies and we want to serve customers first, we make for measure.

A mathematics platform of the future shall provide

interpreters for free-form and syntax-tolerant input
a mathematica programming environment for building domain specific languages
a knowledge-manager that organizes objects, models, methods and implementations orthogonally
a kernel system with a vast variety of symbolic and numerical schemes
technologies for the integration of proprietary solvers
massive curated data in many fields
methods for data mining and machine learning
dynamic visualizing
high performance computing capabilities in the cloud
web connection layers and tools

In the last weeks, I often thought about an internet of finance, an internet of medicine, an internet of environmental sciences, ......
Some of them will need a strict separation of the computational part and the data (like in finance, data nee to be encapsulated at the bank, but all, market observation, valuation and risk-analytics shall come from the plug).

Wolfram|Alpha has revolutionary technologies for the creation of the-internet-of ... and its core technology Mathematica in interplay with proprietary solvers allows for the preparation of the computational foundations.

A question remains to me, where are the visions, business principles, drivers .. that brings us closer to the internet of .. ?
Can we leave behind the thinking in my computer, my software-box, ... Can the makers?

We Speak at Analytics & Risk Technologies in Finance 2010

2010-09-24T06:50:00.000-07:00

4-Nov-10, 10:30 am to 4:30 pm, London.

Finance 2010 Conference organized by Wolfram Research, makers of Mathematica.

In our talks, we want to point out that there are obvious technical and economic reasons to build large scale financial solutions in the framework of Mathematica and how UnRisk ties numerical schemes that are not common in financial circles and symbolic representations to know-how packages. And we want to point out that Mathematica's dynamic visualization techniques enable risk professionals to expose risk that is usually hidden in traditional representations.

Talks will cover

background, technologies and coverage
why instruments-models-methods-implementations shall be organized orthogonally
why advanced numerical schemes and parameter identification integrated into Mathematica matters
why advances visualization techniques unmask risk hidden in plain sight
what implementers can learn from link technologies, HPC computing and the PlayStation

Answering questions like: can model risk be quantified? Is Montecarlo just gambling? Why are trees so bad and Finite Element techniques required? The parameter fit is so good, why is the price so bad?

In the talks, will give exclusive insight with full explanation of the mathematical techniques and present live examples from the advanced bank practice.

Blank Swans and Blank Strips

2010-09-09T01:18:00.000-07:00

Yesterday, I have posted about the book The Blank Swan , Elie Ayache, author, and its challenging thoughts behind and read Jon Mc Loone's post Self Description , which also addressed muscles in my brain it has not had to use for a considerable period of time.
And then I thought, whether the continuous replication of derivative financial instruments that are blank before they get a price on a market and FixedPoint iteration, starting from blank strips, to an amount of black in the image have "something" in common?
(no, it is just a gag - the underlying of the derivative is different from itself, so there is no self-reference; if the derivative is perfectly hedged by the underlying it becomes"redundant", but still have a price on a market :=) )

We Speak at the Wolfram Technology Conference 2010

2010-09-01T02:43:00.000-07:00

We not only attend the Wolfram Technology Conference 2010 but also speak. In "UnRisk - Taming the Machine Infernal" we will highlight some of the new algorithms implemented in UnRisk utilizing the power of GPUs and the application of these algorithms to problems in quantitative finance.

Because Mathematica is used as a universal interface to the GPU programming frameworks CUDA and OpenCL, UnRisk users are not exposed to the low-level complexities of GPU programming.

The talk stands on the stable ground of comprehensive experiments. I have compiled the results in my other blog - Taming The Machine Infernal .

In Champaign we will present the latest results and emphasize on the architectural benefits gained by the usage of Mathematica. It is really amazing how our decision to use Mathematica as platform for UnRisk pays back. Even the most complex components can be tied together so easily. Coarse- and fine-grain parallelization, transaction processing, dynamic visualization, task and work-flow automation, ...

We really look forward to joining Wolfram and other developers and users at the conference.

The Web is Dead. Long Live the Internet

2010-08-25T02:59:00.000-07:00

Wired article, Aug-10 by Christian Anderson and Michael Wolff.
Two decades after its birth, the WWW is in decline, as simpler, sleeker services - think apps - are less about searching and more about getting. I like the structure of this article with the confrontation: Who's to blame: Us - Them.

I have a simple view to this: web browsers "promised" to become little powerful client operating systems, but the service providers' strategies banked on controlling a major chunk of the whole net universe by providing information services with navigation type interaction. From a business stand point innovators required to find other ways so get their market segment. Something alternative, but on the same platform. Web Apps (if you look into my "Music From All Directions" gadget below ....)

What does this mean to computational math? Number crunch in London, manipulate from Singapore? Yes.
If you have made Mathematica applications, you take webMathematica redesign the application a little and make it web enabled - exploiting the dynamic graphics capabilities derived from the powerful Manipulate function. With mobile devices, like the iPad, your application follows your users. The iPad is perfect for graphical apps and wide access to servers.
We have made the UnRisk FACTORY web-enabled, but as a universal solution its front-end has the navigation-paradigm and not an app-type.
But with webMathematica and webUnRisk one can build high-powered quant finance applications with front-ends that contain apps.
Number crunch in London, valuate in Singapore, back test in NY, report in Vienna, ...
I think of Wolfam|Alpha architectures and APIs .....

Artificial Intelligence and Natural Stupidity

2010-08-13T06:25:00.000-07:00

Again, a Blogtitle from Emanuel Derman's Blog. It links to an NY Times article, A First Church of Robotics , by Jaron Lanier.
One of the key characterizations in the article: technology is essentially a form of service (suggesting, IMO, shall be nothing else than ....). We work to make the world better ... and bring more beauty into the world ...
I don't agree with this. I rather see the chance of co-evolution.
Technology might help me to transform knowledge into margins or aesthetics. But it can also drive me to understand the nature of margins and aesthetics better. We might use our brain different, when communicating with another intelligence?
Intelligence is a huge subject, I do not want to discuss here, but Math Plugged touches mathematical intelligence, and why it would be quite poor without computers?

We thought computers are strong in algorithmic math whilst we re unbeatable in axiomatic math. But symbolic computation, automated theorem proving, quantifier elimination, ... show that this is not true any longer. OK, this might be still not be intelligence, associated to the idea of consciousness, ..... but enough to prevent me from doing stupid things (human proofs are often made with a nice taste, and even, if they are flawed, if widely accepted they are pushed into the mathematical knowledge base).

How will I use my brain during my little sessions with Wolfram|Alpha, now and in 5 years?

Forecasting or Predictive Modeling?

2010-07-29T01:10:00.000-07:00

This question is currently spiritedly discussed in the Liked-in group Advanced Business Analytics (just as en example). Predictive modeling is the process of modeling future behavior. The model might be the discovery from a theory, knowledge or extracted from data samples. A predictive model forecasts.
I expect from predictive models that they are interpretable and computational (white box - black box principles drive the spiral of innovation in computation (and math)).

In general, there seem to be uncertainty about the process of generating/manipulating models and their usage. If you want to automate the whole process you need to understand that model generators/manipulators act on (program)trees and the model acts on tuples. Genetic algorithms act on tuples, evolutionary programming acts on program trees.

If you want to do both in one environment you need symbolic and numerical computation, pattern matching, .... Mathematica. (you can implement a genetic algorithm practically in any language, but a general Fit function that is result of an evolutionary selection and combination process (by crossing over subtrees of nested functions and operators) ... ?!)

Spy Stuff

2010-07-09T08:30:00.000-07:00

If I want to provoke mathematicians, I say mathematics is data compression.
But there is more .... Spy Stuff with Mathematica, a great post by Jon McLoone. Hidden in plain sight?

Being Tracked?

2010-07-05T09:30:00.000-07:00

After the contribution about Alphonse Sax .... I remember, we have done some analysis of music, detecting anomalies and assess quality based on audio signals. Because signals have no memory you need to apply and adapt decision rules and this type of automated analysis is error-prone.
In Wired UK ("You are being tracked") I read about developments, triggered by the security circles, called malevolent intent detection, analyzing, say, blood oxygenation and skin temperature, heartbeat, micro expressions of the face, voice, scent, .. and even suspicious kinematic behavior - in order to read indicators of stress and anxiety?!

The Big Brother aspect is one thing that needs to be handled politically, but technically?
I think of the difficulty of distinguishing near-to-spoken singing in an opera and a punk rock sequence .... An error has no consequences for a human.
I know the approach does not seem as stupid as the one of The 7 Deadly Sins, and I repeat, to analyze rough market behavior, say, in finance, might create some additional info for risk management, but I would not knowingly provide mlf to malevolent intent detectors.

You can take this as hippocratic oath ...

1912. Alan Turing Was Born 23-Jul

2010-06-24T05:30:00.002-07:00

First I found this in Wired's This Day in Tech blog.

And remembering the logic lecture at my universitiy, starting with the theory of incompleteness, Kurt Goedel, I thought I write a little .... why this matters .....

But there is a better way. Let Stephen Wolfram speak: Happy Birthday, Alan Turing

Wolfram Technology Conference 2010

2010-06-16T00:06:00.000-07:00

has been announced recently.

Being over 60, it is a bit scaring how time goes by, but we are looking forward to be there and absorb all news and see new developments live. As outlined in my blog description, I am interested in the intersection of math and computer science, but also in domain specific languages, data mining and machine learning, .. and due to the emergence of Wolfram|Alpha I expect answers to questions before I was able to ask them.

Our top talents must be there.

As Gadgets Take Over, Focus Falters?

2010-06-14T06:51:00.000-07:00

this the title of an article in The NY Times, by Matt Richtel. Scientists say bursts of information can undermine our ability to focus. While many people say multitasking makes them more productive, research shows otherwise. Heavy multitaskers have more trouble focussing and shutting out irrelevant information. Although technology users can benefit the brain in some way.

I relate this to decide-in-a-fraction-of-seconds stress, when having only poor information available. See Unstressed In The Air. Decision support requires model and simulation support. Technology development often comes in jumps. And in many fields there are phases where uncertainties dominate comfort. Take automation, bioinformatics, quantitative finance , ... any incomplete decision support (control) need multitaskers to run ...

For reliable decision support you often need to extract models from data, combine them with known theories and do cross-model validation, in line precision control and what have you to analyze features, predict behavior or control processes.

Wolfram|Alpha is not just an incremental innovation. It is radical in the sense that it collects data and applies every known model, method and algorithm and allows for computing whatever can be computed about anything.

Gadgets that run a computational knowledge engine should become un-stressers?

More?

IMO, Wolfram|Alpha has provided an unparalleled framework that allows for a broader coverage of domains, theories and content. Its technology is generic in the sense of data prepararttion an computation (symbolic code in Mathematica, webMathematica and gridMathematica, data analytics, ..)

Recently I read, a typical car contains about 2000 components, 30,000 parts and 10 million lines of software code. I am sure this software complexity is driven by performance demands and safety regulations (with a lot of redundancies). I see a car as a net of sensors and actors. A computational knowledge engine could ....

Domain-Specific Languages or Task-Oriented Languages?

2010-06-01T06:03:00.000-07:00

If you program a robot or discrete manufacturing cell? You do not use C++, C#, Java, Python, ... constructs. Your robot/machine programming language might be implemented in C++, ... but you want to describe "Pick", "Place", "Grasp", "MillPocked", "TurnContour", "DrillHole", .. in operation sequences. They are domain-specific and powerful, if task-oriented.

Domain-specific languages (DSL) are dedicated to a particular problem domain, a particular problem representation (tat can be described in a language) and a particular solution technique (its operational semantics).

Verilog, is an electronic hardware description language, R, for statistical computing and graphics, SQL, the relational data base language, Modelica, the language for modeling physical systems, are examples of DSLs. Not all of them are task-oriented.

Mathematica is a DSL. And MathModelica is one on top of Mathematica. It has Modelica AND Mathematica. UnRisk-Q , the valuation language for derivatives and structured products is a DSL. Mlf is one for the description and operation of machine learning and data mining tasks. Both UnRrisk-Q and mlf combine their specific (task) constructs with Mathematica.

So Mathematica is a kind of mother of DSLs.

My 20 years of Computer Mathematics

2010-05-28T01:26:00.000-07:00

"When I was young, I didn't know any old people. When we did the microprocessor revolution, there was nobody old, nobody. It's weird how old this industry has become", says Bill Gates (highlighted at the article "Master Minds" in Wired, May-10, by Steven Levy - who referred to his book Hackers: Heroes of the ComputerRevolution).

When I was middle-aged, I did not know many young people working in the field of computational mathematics. The symbolic computation community, in the late 1980ies was dominated by middle-aged professors, as well as numerics was. No computational mathematics hackers around.

The field was dominated by the objective to replicate mathematics in computers AND influence programming by mathematical concepts.

When working in factory automation (dealing with geometry, kinematics and dynamics of controlled mechanisms), I felt how important this will become.

And after having met Stephen Wolfram (1990) personally, I knew this will dominate my future technical and business life. Mathematica as the system on the intersection point of mathematics and computer science. Then on version 1 now on version 7 (close to ..).

But conceptual revolutions seem to take longer: I am still astonished about discussions like, why do quants love C++ , or MatLab versus Mathematica?

One of the outcomes of computational mathematics thinking is the distinction between the language of mathematics and its operational semantics. You can wrap any algorithmic representation, implemented in almost any of the major languages with Mathematica and deal with it like it was an original construct in the Mathematica kernel (you extend the language and provide your own operational semantics - in Mathematica, C++, Java, ... and yes, MatLab).

So it needs some patience to convince a broader community that Mathematica is not just another tool set. It is a concept and an integrated approach for accelerate the innovative spiral.

Maybe too much patience for the very young? I am over 60 now, BTW.

Explorative, Constructive Learning

2010-05-07T07:38:00.000-07:00

My job has turned me into a generalist rather than an expert. In the interplay of birds and frogs, I am too much a bird.

Yesterday, I closed a mile-stone sale and just out of curiosity, I declared today's afternoon as learning hours. What would I have done many years ago?

So, I opened a few Wolfram|Alpha windows in my browser and played around first top down asking about certain topics in mathematics.

Then, solving polynomial equations, I played around with degrees and coefficients. I remembered that in a brainteaser forum the probability-a-quadratic-function-has-two-real-roots was asked and I thought how can this be generalized to the general case of degree n (do you need to transform this into questions on probability density functions?).

However, "simulating" lower degree polynomials I "found" clustering of roots when especially playing around with, say "middle coefficients" and they also influence the number of real roots.

So, I needed to look into probability and distributions and played with examples again.

Yes, it is possible to explore and construct knowledge from computations. If you have the right engine that reports not only results but also views ....

I really should take an hour per week at least to repeat such sessions, and not wait for the next mile-stone sale.

Learn The Language of Data?

2010-04-27T03:10:00.000-07:00

Wired, May-10, article by Clive Thomson . Yes, it also can drive me nuts to hear questions like, how can global warming be real, when there is so much snow?, or, why should we face an economic downturn when Apple is doing so fantastic?

This line of reasoning is wrong, because it is about snap shots and not about trends.

To understand trends you should look at massive data and apply statistics and the mother of it probability. C. Thomson claims Statistics should become a core part of general education and summarizes if you can't write you can't think. The same is now true in math. Statistics is the new grammar.

I estimate the value of data driven methods high, but it has its limits and traps. One is generalization. Predictions out side the working space of your samples might become tea-leaf-reading.

If you are lucky and have a theory you have not only correlations but also causation, and if your theory can be described in the Language of Mathematics you are able to solve its models computationally.

BUT, you quite often cannot describe real world phenomena in their whole scope and complexity by mathematical models . Then do not choose either, but combine data and model driven approaches and let a system that speaks the language of data and the language of mathematics do the computational work for you. Mathematica. Your work is to describe the problem and interpret the results.

Co^2-evolution

2010-04-12T07:51:00.000-07:00

Mathematica is the system in which Wolfram|Alpha is implemented. It could not have a more solid foundation.

In my History of Future topic, I touched a few things about declarative programming (Mathematica's unambiguous language) and Quantifier Elimination a fundamentally mathematical technique to answer questions computationally. Is-there ...? yes-it-is-and-it-behaves-like-this....!

Wolfram|Alpha robustly supports free-form linguistic input and serves as knowledge engine. Mathematica's general symbolic language gives the framework of the knowledge representation and its vast built-in solvers provide the computational foundation. I always liked Mathematica-the-platform. And the Wolfram|Alpha Web Service API makes itself also platform. Call it from your programs.

So, a Mathematica program might call W|A that calls Mathematica. This might drive a co-evolution of new requirements and computational knowledge representation.

In How the Tablet Will Change The World (Wired), Steven Levy finds always-on tablet devices will blaze a path of future computing. In fact, they aim to make operating systems nearly invisible, to download and run useful daily apps, but also access power applications in the cloud.

Wolfram|Alpha on iPad let W/A go, where you go and again drives co-evolution of new requirements and .. the co^2-evolution.

Spell The Version Chaos

2010-03-26T02:23:00.000-07:00

Named objects-models-representations-algorithms-implementations-components-variants-revisions.

Sounds as a perfect set-up for chaos?

Take a financial object, like a barrier option (up-and-out, up-and-in, ... down-and-in, double, ..). It might be valued by a Black Scholes model, extended with a volatility surface (Dupire) or stochastic volatility (Heston) and jump diffusion (Bates). All represented by SDE or PDE/PIDE, solved by Montecarlo simulation, (poor) PDE solvers like tree based methods, or better ones with finite elements, improved by upwind techniques, .. in Mathematica, C++, C#, Java, ... improved over time ... heeelllpp?

There is a simple but powerful solution.

Declare objects-models-representations in Mathematica's programming environment, integrate your special solvers into it and manage and deploy projects in the Wolfram Workbench2 that integrates with Wolfram technologies for a complete development solution. This includes hybrid language developments. The Workbench supports Subversion for configuration management.

That completes what we need when developing UnRisk in Mathematica, C++ and Java, managing financial instruments, models, methods and implementations in releases orthogonally.

Not surprisingly, UnRisk 4 (released in Oct-09) is the 18th major release since we have launched UnRisk end of 2001. UnRisk 4.1 to come before summer. Integrating hundred-thousands of lines of Mathematica, Java and C++ code.

And we are a quite small outfit that benefits from a short history of being lucky, namely having selected the most advanced platform at the right time. Wolfram technologies.

Industrial Revolution of Data

2010-03-09T06:52:00.000-08:00

Data-data-everywhere, was a special report on managing information, The Economist , 27-Feb-10. With the-data-deluge on the cover.

In The Petabyte Age , I have provocatively written from mathematics as a data compression field and mentioned new mapping techniques as new ways of visualizing data in I Saw It With My Eyes .

Monstrous amounts of data are produced by machines and the best way to deal with all that information is by machines.

Some say they need watching, I say most probably WE need watching when recognizing pattern and features. At one hand, we tend to oversimplify ( Beware Wind ) or if you torture data long enough often they show you everything you want to recognize.

Hal Varian, Google's chief economist, predicts that the job of statistician will become the "sexiest" around. Machine learning expert, I add.

If we want users to find the needle in the haystack, you need to create information on information, if you want to recognize structure in information you need to create models from data. Lucky if they are computational. Even more lucky if they are computational and interpretable, as most of our mlf models .

20 great foods I am not eating

2010-03-03T05:31:00.000-08:00

The easy-to-buy-superfoods could help me to live a healthier, flat-bellied and longer life (TimesOnline).

Yes oily fish, new potatoes, olives, .. are great but in a meal often it is not the key ingredients, it is the configuration, process and work flow, the non-stick pan, the resting of meat, the relaxing of the fish, the temperatures, the timing, the sequencing, the seasoning, the semi finished ingredient making, like the stocks, ... that makes a meal not only tasty, but healthy.

I like Gordon Ramsay's cooking and his Chef's Tips (like in TimesOnline) for process-through consistency. Or Heston Blumenthal, Massimiliano Alajmo, Alain Ducasse, who also have clear cooking principles

In mathematics, we also have special functions and semi finished configurations. We might find some of them more tasty and healthy, but we want to cook them to tasty models and systems, all in a consistent process and work-flow, relying on a growing mathematical knowledge base.

Mathematical foods I am not eating? Libraries for example.

Mathematica allows me to do mathematical cooking consistently. It is a computing environment, a language, a vast knowledge base and comes with a workbench. Thousands of functions, are consistently designed to be nested and work together. It automatically selects algorithms and tunes precision. It allows me to integrate my own food.

From a 10 line program to a million-line production system. For tasty and healthy results.

Closer To The Internet Of Finance?

2010-02-22T08:58:00.000-08:00

In previous posts I have already referred to Emanuel Derman's Blog (one of the most influencing quants, IMO). In his latest post Modern Portfolio Theory And Its Effects On What You Cannot Buy he mentioned a search in prominent sites with financial data.

And some disappointment about the incompleteness of such information, missing practical things, like volatilities of a stock, its correlation with ....

And some surprise that he found a useful set of such information and visualization in Wolfram|Alpha, simply entering the stock symbol. Like AAPL ( equal to Apple stock )

And I think of decision support for global risk management, often requested by regulators. But how to do?

Building an internet-of-finance with massive data management and analytics management on a computational knowledge engine. Wolfram|Alpha.

Hybrid Programming made Easy

2010-02-22T05:05:00.000-08:00

Never Work Without a Workbench: Our UnRisk integrates hundred thousands of lines of integrated C++, Java and Mathematica code with data base access, number crunching in Grids, deployed in the Web.

The Wolfram Workbench is perfect for that.

No surprise that it is the integrated development environment for the huge Wolfram/Alpha system.

Wolfram Workbench 2 has been just released. We are very much looking forward to exploit its new features in

Authoring and document development
Installation manager
Front-end integration
Eclipse update

It really rationalizes our development process and standardizes the work flow in a multi developer - multi technology - multi method environment.

Losing Big - A Winning Strategy?

2010-02-01T02:43:00.000-08:00

This is suggested by the Title Story , Wired Magazine Jan-10 (indeed, an interesting view; I especially like the The-Fall&Rise-of-Alec-Baldwin part ;) ).

I doubt that the financial crisis is result of such a strategy, but let us hope that the conclusions are. Its greatest mistakes, IMO, public intervention, leverage because of highly correlated deals, too less cost of money, toxic instruments - giving the false comfort of bullet-proof vests, but also the race of model complexity (I wrote about this in When Good ENUF is Great ).

However, is it necessary to train our brain to embrace failure, to open ourselves to new discoveries (Wired) by suffering from real downturns? As neuroscientists suggest. If we don't "feel" the failure, we ignore it as anomaly of the experiment? Maybe true.

But often real live downturn experiences took too short or too long, or destroy everything, ...

What about provoking mistakes in scenarios, make stress tests on pathologic cases and demonstrate bad behavior in simulations?

Reinvent Development? My 10 Ideas for 2010.

2010-01-25T06:03:00.000-08:00

Reacting to new market regimes. Reinventing our development? Ideas ranging from further productivity boosting to closing customer feed back loops.

1. Progress motivates. Developers working out evolutionary development cycles will have more best-workdays.

2. Technology revolutionizes workflows. Requires? Integrated model builders and quick demonstrators. A declarative programming environment. Document centered interfaces and several deployment options.

3. A bank of innovation. Establish a lab to design, analyze and "fund" innovations and boost dynamism by requiring alternative methods, models and solutions.

4. Get the tools required. Enhance our platform and workbench for hybrid, multi-langiage system development.

5. A fast path from lab to market. Remove deployment obstacles and close the customer feed back loop.

6. Support rule breakers. Standardization is good, but often new methods and tools will more empowering our developments.

7. Spotting errors in early stages. Needs automated precision control and analytics and ease of cross model testing.

8. Development cities. Teams of developers who do everything from design-to-testing, deployment and support.

9. Independent from usual roles. Why pretend that only sales shape customer relations? Developers daily interact with actors and communicate via the underlying technology.

10. Inside-out knowledge. Developers not only provide product use training, but also seminars giving full explanation on theories, models and methods.

Mathematica is our strategic platform.

The Sky Falls? The Sky Falls?

2010-01-18T04:32:00.000-08:00

From Wolfram Blog rain in England around Swithun's day .

St. Swithun’s day if thou dost rain
For forty days it will remain
St. Swithun’s day if thou be fair
For forty days ’twill rain no more

Jon Mac Loone from Wolfram Reasearch, International Strategic Development tested, whether this is "true".

If you would bet on reverse rain floaters? A portfolio built of locations?

Inspired by an article of Aaron Brown in Wilmott magazine, Jul-09:

In financial risk management you might calculate VaR that, simplified, tells you, that there is a 1% chance your trading position would lose more than VaR (the days you get really wet in double sense).
But this is not enough. Risk managers should perform backtests like

the actual fraction of VaR break days is 1% within statistical tolerance?
the VaR breaks are randomly distributed (avoid Swithun situations)?
the VaR breaks are independent of the level of VaR?

If you have the VaR, which calculation can become nasty with the most sophisticated deal types (you can do it in UnRisk), back testing is as simple as in Jon’s example. In Mathematica..

2010, Year of Dekonstruction and Reconstruction?

2010-01-07T06:37:00.000-08:00

Ferran Adria the celebrated chef of "El Bulli", deconstructs and reconstructs, say, fish, fungi, ..., fragmented photography does and TRIZ innovation principles suggest a kind-of for re-inventing products.

In Asymptotic Mathematics I introduced an example in which mathematical solvers do. Decomposing domains as fine as exact solutions can be obtained and aggregate results.

In reconstruction you might want to use your exact solvers as base for simulations along a wide working-space and produce i/o samples that you can analyze by adequate machine learning methods. Extracting ultrafast Surrogate Models .

In finance, climate, biology, .. we need to model complex behavior knowing that more parameters and complexer models might lack the robustness required (you might get lost in the numerical jungle, stuck in local minima, inverse problems, ..).

Mathematica and mlf enable us to make comprehensive but low-effort experiments with deconstruction and reconstruction in quant finance.

This motivated me to playfully look back at the past from a future perspective.

Me talking to Myself in the Future

Geologists Love Beer

2009-12-18T05:19:00.000-08:00

Why? explained a Wired article today (with winking eyes).

It has to do with the amount of time spent outside doing fieldwork; with the many places where they do not have drinkable water. Or - my favorite: when it is hot, and you've been hiking all day carrying 50 pounds of rocks, do you want a Merlot?

But: science does not work, when people keep secrets and don't share their data (and what could better help with free flow of information).

This seems to be in a little conflict with How to Pick a Good Fight . And free flowing beer for free flowing information seems to be ineffective for a good future-facing fight?

Mathematica's declarative environment and document centered approach helps with free flow of information. But it is also a great platform for good fights. Mine innovative solvers by attacking problems with different approaches.

And you get clean Mathematica in most of the places.

Silo but deadly?

2009-12-14T02:37:00.000-08:00

The title of an article in The Economist (subtitle: messy IT systems are neglecting aspects of the financial crisis) suggests that many banks have huge problems with data quality and the fragmented IT landscape made it exceedingly difficult to track a bank’s overall risk exposure before and during the crisis. But also concluding answering such questions as ‘What is my exposure to this counterparty?’ should take minutes. But it often took hours, if not days.

The core question in difficult times: shall we emphasize on the technological support of our core business, or optimize the work flows? People often think, there is a trade off between supporting quickly emerging business ideas (in finance: new deal types, tradig strategies, ..) or the organisational aspects (including risk management).

It is assumed that mapping the quickly-emerging lead to quick-and-dirty prototyping that leads to quick-and-dirty sub systems.

But THERE ARE evolutionary prototyping platforms.

The compute-develop-deploy paradigm of Mathematica is a leading example, for all developments in quantitative fields, from the user-programmed desk top application to the largest scale enterprize-wide risk management system. With its link technologies Mathematica can integrate even legacy-system fragments, and its data base link allows for high data quality and consistency. Intelligent data aggregation and scalable built-in high-performance-computing enables in-time-decision support.

Again, it is not OR, it is AND.

Hopenhagen

2009-12-09T04:17:00.000-08:00

Some thoughts on the big climate conference in Copenhagen. I am not a climate expert, not in social and economic sciences and not a politician. But interested in complex-systems theories, I have reservations about the linearity of the discussion. We will have deadly global warming and we can (only) avoid it by CO_2 avoidance (entering the anti-Carbon age immediately)? Or, we have nothing of that?

Assuming predictive modelling has produced correct results; how can we exactly know, where to put our concrete efforts and money? In the extreme: put them into hectically reducing CO_2 or measures to avoid negative impacts on people, social entities, the community, if "it" happens? Wouldn't our civilisation change in an atmosphere of hysterie? Do we know the social impact and cost?

The co-evolution of greed and fear has no equilibriums.

However, something will happen, consequently, something must happen. Crazy? Not at all, IMO.

It is always a reasonable objective to make our environment cleaner and save resources.

And probably a radical innovation was to exploit, say, weather and climate drifts and volatilities for energy and process optimization? This might lead to new systems of "technical" co-evolution (work with the knowledge about the partner system, to optimize the own)?

Future-Prize to Fraunhofer ISE

2009-12-02T23:53:00.000-08:00

The 2009 Future-Price of the German Federal President (Horst Koehler) went to Fraunhofer ISE (Institute for Solar Energy) and BASF. Small spheres against climate change - Energy efficiency and comfort thanks to intelligent building materials. This was celebrated yesterday in a show at a major German TV station (ZDF).

Congratulations!

Fraunhofer uses Mathematica since 1992.

Fraunhofer ISE has the largest user group within Fraunhofer and hosts one of the yearly Fraunhofer Mathematica Seminar.

Huginn and Muninn

2009-12-02T03:39:00.000-08:00

the pair of ravens that flew all over the world, Midgard, and brought Odin, head of the Aesir, information (attested in the Edda, written nordic mythology). Some interprete them as ear-whisperers, some as the personification of the god's intellectual power (Huginn - thought; Muninn - mind).

However Odin worries about the return of them each day and whether he can transform information into wisdom. He also worried whether this was enough and gave one eye to get a drink from Mimsbrunn (Mimir's well), to know yet more.

We don't need to give one eye and also ravens are replaced by massive information available online. But still we need to decide how we combine modeling and data-driven methods for decision support.

We have integrated our multi-method machine learning methods into Mathematica to create computational knowledge from modeling and data mining. All implemented in machine learning framework (now on version 2), for better decisions. In business and industry. We want customers that want to turn information into values. Unlike Odin, who wanted to become all-knowing and all-mighty and finally led Asgard into Ragnarök (end of ruling powers).

Must Climb The Tree To Get The Fruit?

2009-11-30T00:55:00.000-08:00

(Photo from FreeDigitalImage ). In the development of software-development-paradigms you at least implicitly find this suggestion. In any compile-build-go cycle, architectural design, algorithm design, data-structure design,... induce the quality of your system, especially with quality features, like robustness, extendability, maintainability, ... (short, software life cycle planning). Even if you do explorative and experimental prototyping, you feel you must climb the design tree to get the fruit of good runtime software.

But there are fruits on pre-designed plants (grapes, ..), only mans-high.

If you Develop in Mathematica you inherit a lot of things, from an open architecture, link technologies, a language for all programming paradigms, especially literate programming, .. to a development environment that supports you when doing personal to large-scale-hybrid-system development (the Wolfram Workbench , for desktop-, grid- or web applications).

And it changes the development principle to Compute-Develop-Deploy. Pick the grapes from a well-designed wine-plant that you have tuned to transform the strength of your soil (terroir) into an outstanding taste (for immediate eating or wine making). Configure your plantation ...

You don't need to climb the tree all the time.

What Would Newton Do?

2009-11-24T05:49:00.000-08:00

After the 3 Mathematica seminars for Fraunhofer Gesellschaft I thought: with over 60 institutes it carries out research in hundreds of technology fields. With this coverage a broad variety of theories, models and algorithms are found, verified and validated in Industrial Research projects. How to verify and integrate this knowledge?

Isaac Newton comes in my mind, one of the most influential men in history. I do not need to repeat the broad coverage of his ground-breaking discoveries (from fundamental laws to concrete equipment building). One of the last universal scientists? His experiments were a means of verifying (not discovering) what he already knew (from a speech about him). Or, he found a proof in days for what he knew for years. An experimental natural philosopher?

In the discussion with the Fraunhofer researchers it often turned into, "how can we combine our modeling and simulation systems with the laboratory systems?" It is about, different paradigms, system breaks and the question whether hybrid systems can work?

Yes it can, Mathematica, with its multi-language link technologies and the ability for high-level declarative task programming is the platform that can tie technologies together.

Newton WITH Mathematica. Was he then more an explorative, constructive natural philosopher?

Final In The Castle

2009-11-22T01:10:00.000-08:00

20-Nov-09, FIT, Institute for Applied Information Technology, organized the final-of-three Mathematica 7 compact seminars.

The Fraunhofer institute pursues a user-centered approach to information and cooperation systems design. Usability and usefulness of IT are optimized in the interplay of work practice, organization, and process design.

Harald Mathis, Head of BIOMOS, Biomolecular Optical Systems, provided the most beautiful seminar room in the castle of Birlinghoven / St. Augustin. Thank You!

Due to the IT focus of most of the attendees, Oliver Ruebenkoening emphasized on the platform character of Matematica 7 and gave full explanation on the most important link technologies.

In the spirited discussion this was also the major point: how easy Mathematica can be linked to other systems, like lab systems, data analytics systems, or control systems. Mathematica's declarative programming environment is perfect for coordinating and controlling them and apply task automation.

To me, it is always impressive, sitting in the train from Frankfurt to Siegburg/Bonn, browsing through the tunnels with over 300 km per hour. A warm! sunny Nov-day and a walk through the spacious park and woods around the castle added to my excitement.

Sun IS Shining

2009-11-15T02:36:00.000-08:00

13-Nov-09, ISE, Institute for Solar Energy Systems , Freiburg hosted the second-of-three Mathematica 7 compact seminars.

The Fraunhofer Institute develops systems, components, materials and processes in the areas of the thermal use of solar energy, solar building, solar cells, electrical power supplies, chemical energy conversion, energy storage and the rational use of energy.

Axel Brendle, Head of IT, set up a perfect seminar environment in ISE,s impressive building. Thank you!

900 persons are employed at ISE. They have access to the latest technologies for their research and ground-breaking technology development. Networked and grid-enabled Mathematica 7 is widely used in modeling and simulation.

Oliver Rübenkoenig gave his talk for over 40 researchers from 10 Fraunhofer institutes located in the southern region of Germany. He emphasized on the fact that Mathematica's technology-foundation ties into many new features in Mathematica 7 and finally WolframAlpha for application developers.

The spirited discussion showed that computing-developing and deploying applications are increasingly important in industrial research. As an example, I want to mention ITWM's world-wide released Analog Insydes .

Again, I approached Freiburg by train from Frankfurt browsing along the Schwarzwald with the vineyards of famous vintners on the foothills (like Berhard Huber, Malterdingen, and his fantastic but rare Spätburgunder "R"). Innovative as Fraunhofer, in his field. And I like to walk through Freiburg, even if I only have an hour.

I am looking forward to the seminar in St. Augustin/Schloss Birlinghoven.

Those Are The Days

2009-11-08T01:31:00.000-08:00

6-Nov-09, Fraunhofer Institute IWS, Dresden: the first-of-three Mathematica 7 compact seminars, a joint event of Fraunhofer, Wolfram Research and my uni software plus, took place at the impressive IWS facilities.

I found it exciting, and, I believe, also the participants from ten Fraunhofer institutes.

Oliver Ruebenkoenig, Mathematica Kernel Developer, the presenter, managed to give a guided tour through the most beautiful areas of Mathematica. It is not so easy to walk through a platform that has many thousands of functions transformed into ground-breaking technologies in 2 hours; he chose a perfect balance of in-depth sessions giving full explanation on capabilities and implementations of Mathematica 7, a huge release, with overviews, always giving the link to the rich resources in the comprehensive Wolfram pages-for-users. From dynamic graphics examples to Wolfram Alpha. We had a spirited discussion with the attending Fraunhofer Researchers, from the first-viewers to the advanced mathematica users.

Andreas Kluge, IWS IT, organized the event in a great venue. It starts with the entry, with presentations of the institute's achievents and goes to a perfectly equipped seminar room. Thank you!

But I even got some extras. When I approached and left Dresden, a glistening sun made the colored leaves in the woods golden surfaces and it was the first time that I took time to enjoying a walk through the impressive Altstad and Neustadt of Dresden.

Im a very much looking forward to the next Fraunhofer Mathematica events in Freibug and St Augustin.

Technological Convertibles?

2009-10-27T06:27:00.000-07:00

Sascha Kratky, our Senior SW Engineer, just returned from the Mathematica User Conference. After his summary in a short workshop, I feel a bit dizzy. So many advances from tying things together which have been designed and implemented in a unified way. Things are not reused in the traditional way, they are built on-top-of-themselves and provide the capabilities of, what I would call convertibles.

For my uni software plus , developers and business developers, it is difficult to keeping the pace with Wolfram's achievements in an innovative spirals that accelerates. New possible platforms, solutions, uses and users are just emerging from a fast growing technology base that is convertible.

Mathematica does not only develop in depth but it approaches a universality stage that is unprecedented in technical computing. It drives WolframAlpha widely accessible but with the API you can use WA from your desktop application.

It is close to 20 years that we have selected Mathematica as our platform for creating a new kind of business.

We have recently released mlf 2 and we will ship UnRisk 4 from tomorrow, and after a week UnRisk FACTORY 2, all grid-enabled, task-automating, multi-method, multi-language, multi-front-end. With Mathematica 7 as platform.

And shhh we are exited about what comes next.

My Programs Ask

2009-10-21T00:41:00.000-07:00

In my History of Future topic, I have tried to make evident that my mathematics-related business life was challenged by fundamental technology developments, most now realized in Mathematica that drives the computational part of WolframAlpha the computational knowledge engine.

Not so surprisingly there is an iPhone App .

But the innovative spiral gets another drift by the Wolfram announcements of the Wolfram Alpha Web Service API , giving access to the WolframAlpha platform.

To me this means that programs can ask WolframAlpha and integrate own results into the WA result styles.

Like courseware for Go to School but not Into School or creating kind of EveryWare in any field of computational applications.

It is fantastic. Our senior software engineer Sascha Kratky will get more insight from tomorrow on, at the International Mathematica User Conference 2009.

As Stephen Wolfram says in his blog, A Big Week for WolframAlpha .

The 7 Deadly Sins

2009-10-13T01:53:00.000-07:00

In WIRED, Sep-09, they presented an infographics example by mapping sins created by plotting per-capita-statistics onto the US map. Greed by income spreads, Envy by thefts, Wrath by violent crime, Sloth, Gluttony, Lust, Pride also on single factor stats (no sorry, Pride as overlay of the other). A nice try, but a simplification that I find poor (to say the most polite ...).

In quantitative finance, we would be lucky to be able to model greed and fear, important factors for high leveraging deal spaces and finally bubble building? But their interdependence has its dynamic in a kind of co-evolution. And greed might be good in some cases (reasonable risk taking)? And it is absolute. But envy is bad, because it is relative (to your"neighbour"). Too much envy in finance?

However, the analysis of such "irrational", "behavioral" factors might become more important. But if, we need to put all our best efforts to analyse them in multi-variant parameter spaces and apply multi-method systems, like mlf on Mathematica .

And I think, Stephen Wolfram's NKS !

Unrest before a new UnRisk ....

2009-10-05T08:14:00.000-07:00

I have pre-announced UnRisk PRICING ENGINE version 4 in UnRisk Insight, titled Unrest . Testresults suggest the ralease in about 2 weeks. And we will release UnRisk FACTORY version 2 within a few days after UnRisk 4.

My Unrest comes from my responsibility to do the create awareness actions right. The new releases contain so many new features, but I want to avoid the eager-seller trap ...

In addition to many new features UnRisk 4 and UnRisk FACTORY 2 are full Mathematica 7 versions.

They use its new built-in parallelism, built-in integration tools and also its fantastic documentation framework for in-product and on-the web documentation.

But let me come back to testing.

Our FACTORY combines a grid-enabled PRICING ENGINE with a data base, a web connection layer and comprehensive UnRisk services realized in Java.

Each user-session (automated by scripts) creates a log document in form of a Mathematica 7 notebook that contains each single system-action, automatically. This notebook is re-fed into the PRICING ENGINE and allows us for comprehensive automated test cycles for interactive sessions and batch processes.

The same mechanism enables us to perform unexpected short support cycles.

Having selected Mathematica pays back manifold. UnRisk 4 will be the 16th major release in 7 years. And I need to recognize that excitement and unrest come as twins.

Predicting The Present

2009-09-14T07:00:00.000-07:00

Weather forecasting is a field where advanced modeling is fruit of challenging research from model physics (thermo- and fluid dynamics, ..) to atmospheric chemistry. Advanced numerical schemes and sw architectures allowing HPC and the management of massive data are applied. The approach is multi-strategy and multi-methods. The quality-at-time-in-advance of the forecast is related to the resolution of information.

For many applications highly precise weather predictions are important. Take the energy efficient control of (large) buildings. Focussing on such applications one needs a kind of embedded calibration system which takes the sensor information of the buildings and the coarser weather models and refine them by turning historical informations into precise local predictions. For the next hours.

Blue Sky Weather Analytics, an Austrian specialist for local weather-related products, has developed such a high-precision short term predictor based on our machine learning framework .

Go To School - But Not Into School

2009-09-07T08:02:00.000-07:00

It was a beautiful, sunny late-summer weekend. With a few scattered clouds. I enjoyed it, reading about various things on my terrace.

Buildings in the neighbourhood, even being invariant in their facades, looked individual with unexpected contrasts, in the sun. I took a book on complexity-or-minimalism in architecture ....

and then some articles on complexity-or-minimalism of financial modelling, .... In the crisis, we have learned about many things in the laboratory of reality of financial markets. Shall we continue to believe in the Efficient Market Hypothesis, or change to more evolutionary modeling, simulating the behaviour of market participants?

How many further experiments in the laboratory of reality do we need?

And I refreshed my little knowledge in explorative and constructive learning. Thinking of obtaining computational power, Mathematica and comprehensive courseware from the cloud and use it for the construction of individual knowledge. Life-long.

How long will it take to de-monopolize education from location (school), person (teacher) and content (syllabus)? Forever?

How many cheeses?

2009-08-19T05:12:00.000-07:00

I looked into the program of Mathematica User Cónference 2009 and when reading Wine-and-Cheese Reception I thought: "how many cheeses will people know in different territories in average?" In some, probably 7? I think, I know approx. 100, preferably French, Italian, English, Spanish, Swiss .... with very local ones like Ami-du-Chambertin, Castelmagno, Oxford Blue, Cabrales, Sbrinz, ..

Then I typed into Wolfram Alpha "cheese", knowing this is asking for static info and not computational knowledge. But yes, I got a list of more general cheese types (40?). Not bad, I thought and then "Roquefort", "Limburger", "Gruyere", ... returning useful info, like nutrition facts, nutrients compared to other food, calories, vitamins, ... Wow.

Beware Wind!

2009-08-13T04:57:00.000-07:00

A little more finance. Again from Derman's Blog. In the following paper it is described that studies have found evidence that wind speed has strong influence on mood and consequently effect in stock returns. All extracted from data of wind speed and daily stock market returns across 18 European countries from 1994 to 2004. The authors claim: our findings contradict the rational asset-pricing hypothesis ...

This is a joke, or weather forecasters will become immensely rich and will only publish fake forecasts in the future to avoid sharing.

I really like data-driven methods, but this example shows, how important it is to analyse the full range of factors that might influence and apply multi-method analysis and how dangerous a naive approach can be.

Humans WANT to see patterns and machine learning shall help us to see the right ones. It often requires comprehensive cross-model checks, sub domain analysis, .... This is why we have passionately developed multiple methods and task builders for mlf .

I Saw It With My Eyes

2009-08-12T08:33:00.000-07:00

Our machine learning framework is a multi-method-multi-strategy system. One of the methods: self organizing maps (SOM). A SOM is a type of artificial neural network that is trained using unsupervised learning to produce a, say, two-dimensional map from a usually high-dimensional input space of the training samples. They use a neighbourhood function to preserve the topological properties of the input space. This makes SOM useful for visualizing low-dimensional views of high-dimensional parameter spaces.

In the above maps you see the intensity of missing dots on printed paper (right side) and the values of a special parameter of the printing process (left). One sees that there is a region of correlation in the left/down corner but the many-missing-dots region in the center has no correspondence at this parameter (it might at another process parameter). One could imagine to have a monitoring board where all process parameters and quality parameters are represented as SOM und you control your process by staying in the high-quality regions of your quality parameters.

Surrogate Models

2009-08-06T08:37:00.000-07:00

Remember the HPC post? Simulating a portfolio of financial instruments across scenarios might take days on single processors. What you have to do is solving many millions of complex differential or integro-differential equations. Mathematica helps us to exploit multi-core environments, but to get results in real-time, you need more. In certain cases (VaR), we gain enormous speed-up by principle component application, or, if Montecarlo simulation is required, we put valuations on GPU co-processors.

But we are also studying the possibility to use Surrogate Models (SuM). SuMs are easy-to-compute compact models that mimic the complex behaviour of the models in special ranges used in a simulation. Surrogate models are constructed using data-driven approaches as we know from machine learning in our mlf . Simplified: run the simulation with the "original" models, trying to cover the whole working-space, sample the results and extract the SUMs from them.

The finding process is time consuming, because you need to do massive model and cross-model testing.

In Mathematica's declarative environment it is easy to define higher level tasks, as in mlf, where one can define such machine learning tasks, which then run automatically.

In UnRisk this is an ongoing project, but if we think of principle component analysis, GPU co-processing, SuM and grid computing, we might gain gigantic speed-ups. With the indispensable support of Mathematica .