Unstressed in the Air?

In the golden age of aviation an unprecedented growth in affordable light aircraft occurred, providing ambitious individuals with the means to conquer the sky. Aviators raced to be first from A to B, areal daredevils entertained through flying air shows. They flew in, say, biplanes with open cockpits and most probably controlled flights by watching their vectorising shawls ....

Later, cockpits were closed and pilots got a few instruments and when jet engines replaced classical engines they felt an unknown stress of decisions-in-the-fraction-of-a-second required (beyond the intuitive steering). It is my strong believe, that the stress-intensity held until the auto-pilot were introduced and overtook the control.

However, we know from the Hudson-river-plane-landing that expert pilots need to overrule the auto-pilot, in emergency cases.

Mathematica's core for the mathematical auto pilot is a combination of automated algorithm selection and precision control and tracking. Mathematica looks for you into te problem description, say, a Differential Equation, and automatically selects a solver and tracks the required precision alongside the process structure.

But if you know, your DE needs a special scheme, you can select it from a broad variety of adequate solvers.

Algorithmic Swimming

Interested in quantitative finance, I frequently read Emanuel Derman's Blog . He is one of the most influencing quants. At this time he wrote about algorithmic sports, which catalysed this:

The world swimming championships take place in Rome now. People say, records fall not only because of better training, improvements in the human minds and bodies, .... but better equipment. In swimming? 20 years ago we might have smiled politely, if somebody had come up with such a prediction.

Swimming is a sport that is pretty pure. As mathematics is a field that was always understood as pretty pure. 20 years ago most might have smiled ....

Mathematics plugged, enables radically new approaches like explorative, constructive self paced learning and mathematical problem solving? Breaking records, ...

Mathematica, the mathematical suit that not only makes you much faster because of its structure, but also makes YOU faster in doing math without it. Because your training is much better.

In sports functionaries might find ways to ban the structured swim suits and swimmers will find new ways to swim faster without it - like swimming under water in back stroke .. but one cannot ban mathematical software, because we do now algorithmic mathematics.

I wrote about constructive learning in Courseware Builder .

Complex Bounded Geometry

In computer aided manufacturing you often deal with sculptured surfaces. In metal forming, milling, plastic injection, pressure casting, .... If you wanted to have the perfectly smooth (car bodies, ..) then a Bezier representation is a good choice, or in case of really sculptured, then you might want to deal with B-Spline representations.

But in general, modeling solid 3D objects, which shall be fabricated, the paradigm of choice is bounded geometry and, for mastering the implementation complexity, NURBS are an appropriate unified representation.

In my factory automation time, my team had to deal with such requirements daily, suffering from a Babylonian variety of descriptions, representations and implementations.

With this background it is great o see, that Mathematica 7 offers Comprehensive Spline Support .

Geometric modelers, CAD/CAM systems have this all?

This is correct, but in black-boxes, you have to rust on their accuracy and robustness, or not.

Doing principal design in Mathematica you can get insight, with symbolic handling and in combination with other mathematical settings, dynamic and interactive manipulation, ... Bezier, B-Spline, NURBS, .. "friends" coming back to me in new clothes.

Never Work Without a Workbench.

We develop UnRisk a software solution that uses Java based user interfaces or servelet-based web interfaces on the front-end and Mathematica as the grid-enabled platform for the computational engines on the back end. A system with hundred thousands of lines of Java and Mathematica code. We are using Wolfram Workbench, which has become quickly an indispensable environment, we could not live without. We can develop both Java and Mathematica code in one IDE and profit from the version control and build management infrastructure we had established around Eclipse. We also enjoy the project-based approach, profiling, debugging, .. but especially the gridMathematica support.

How Much Water .... ?

How much water will we have tomorrow? This is the question Verbund experts are asking to optimize their supply and trading strategies. Verbund - Austria's leading electricity producer and supplier operates over 90 hydropower plants with a combined maximum capacity of over 6K megawatts and an average annual generation of approx. 22.8 Billion kWh.

Production output is directly related to the amount of water, hence the headquarter of Verbund does predictive modeling on the availability of water. Predictions relies on a suite of methods but dominantly on neural networks.

Verbund selected the machine learning framework, because of its high-quality neural networks, which can be easily combined with other mlf methods for predictive modeling in Mathematica's declarative programming environment. The Verbund experts also find Mathematica's capabilities of importing massive date important as well as its vast variety of visualization techniques.

5000 Demonstrations

Wolfram has passed 5000 demonstrations published by the Wolfram Demonstrations Project. Congratulations!

Conrad Wolfram has announced this in Wolfram Blog .

I often visit this site myself. My interest is broad, but especially I browse application-oriented demonstrations (business, engineering, systems, ..).

As UnRisk developers, we use Manipulate quite extensively. How? In quantitive finance tasks can become complex and bear the danger of model and method risk that can become horrible in interplay. We test our models, algorithms and calibration schemes across comprehensive scenarios, not forgetting the pathologic cases. In the above visualization you just see a single spot of such a dynamic visualization; the price and the vega of a complex convertible bond over the price of the underlying and time to maturity. If you do not solve the related PDE correctly and robust you get severe mis-pricing and misleading hedging parameters. 

The significance is new insight without additional effort

Brainteaser Killer?

In Wilmott (serving the quantitative finance) you find a brainteaser forum. Due to the topic, problems are often related to stochastics. I draw just one of the recent questions: if you place a normal pack of cards (ie. 52 cards with no jokers) face down on a table and begin to turn over the cards (from the top) one at a time, calling out Ace when the first card is flipped, Two when the second card is flipped etc. right through to King (all irrespective of suit), and then repeating this four times, what is the chance of getting through the whole pack without calling a matching card?

Marco Carreira frequently gives explanatory answers in form of Mathematica programs.

called = Map[Mod[#, 13] &, Range[0, 51]]

{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 0, 1, 2, 3, 4, 5, 6, 7, 8, \ 9, 10, 11, 12, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 0, 1, 2, 3, \ 4, 5, 6, 7, 8, 9, 10, 11, 12}

nrdraws = 1000000;

draws = Table[Map[Mod[#, 13] &, RandomSample[Range[0, 51], 52]], {nrdraws}];

matches = Table[MapThread[Boole[Equal[#1, #2]] &, {called, draws[[j]]}], {j, 1, nrdraws}];

Sort[Tally[Map[Total, matches]]]

{{0, 16175}, {1, 68929}, {2, 143830}, {3, 198761}, {4, 201272}, {5, 161134}, {6, 105438}, {7, 58535}, {8, 28166}, {9, 11655}, {10, 4119}, {11, 1394}, {12, 449}, {13, 108}, {14, 27}, {15, 7}, {16, 1}}

N[16175/nrdraws]

0.016175

He simulates 100000 random draw-cycles and counts the matches results (0, 1, 2, ... matches). I realyl like this. Describe the problem, instead of brooding on an exact solution. Mathematica is perfect for this.

EveryWare - UnRisk an Example

In the mid 90ies, being lucky, we were asked to develop some tools fo some exotic convertible bonds for a London based trading desk. How we did it and have now UnRisk FACTORY an EveryWare for risk analytics.

1. Optimize the number crunching

In a risk management workflow you are crunching numbers millions of times. Pricing and calibration engines are optimized in C++.

2. Provide declarative programming

To described financial objects in the language of mathematics, we integrated our engines into Mathematica and inherit a vast variety of technologies.

3. Parallelize it

We multiply performance by applying coarse grain symbolic parallelization techniques from Mathematica. In a few lines of code UnRisk became gridUnRisk.

4. Make data persistent

UnRisk's I/O relations and and control objects are represented in MS SQL. Mathematica's JDBC support and Java was the natural choice.

5. Web it

UnRisk interaction patterns, represented as JSPs, part of a Java implementation are wrapped up as Servelet web application.

6. Tie everything together

Web browser - UnRisk servelet - data base - UnRisk service - gridUnRisk - numerical engines; using a mix of Mathematica link technologies and communication protocols.

7. Scale it

Scalability is "outsourced" to license management.

Web browsers on smart phones and cloud computing makes UnRisk FACTORY a pervasive computing solution.

All Austrian Universities

Austrian universities use Mathematica since its release with growing intensity.

A few months ago, all Austrian universities and selected colleges entered into a contract enabling 10.000 researches and 150.000 students unlimited access to Mathematica.

The project was conducted by the Austrian ministry for science and research and computing centers of the academic institutions.

Mathematica supports researchers exploring, say, the radiation in chemical reactors, risk of complex financial systems, complicated bone connections, the interplay in social structures or algorithmic composition of music. Educators to give full explanation following the latest didactic concepts. Students constructing, exploring an strengthening knowledge from interactive usage, self paced and often distance.

Based on its document-centered architecture Mathematica stimulates inter-department and inter-disciplinary collaboration.

And academic institutions can run heavy calculations and simulations on their high performance clusters, by exploiting Mathematica 7's built-in parallel computing capabilities.

I am passionate about new technologies, but I like as much to finding cooperative models for the benefit of various groups. This one ...

Fraunhofer, 60 years Anniversary.

60 years of dedication to the future. As it stands today, Fraunhofer-Geselschaft attained a size and influence that makes it an undisputedly vital element in Germany's industrial and scientific landscape. It has close to 60 individual institutes and carries out research in hundreds of technology fields making the results available particularly in the form of research projects commissioned by industry.

In 1992, Fraunhofer has designated Mathematica as its strategic technical computing software. From biomedical engineering to industrial mathematics. With their broad coverage of research fields Fraunhofer scientists enjoy Mathematica's broad variety of algorithms with integrated symbolic and numerical computation, the management of massive scientific data, high performance computing all integrated into a highly-productive development environment with hybrid-language interaction, system integration support and innovative deployment options based on structured interactive documents or the web. Unlike other systems, Mathematica is from-one-casting.

Wolfram's flexible general license contracts made it easy to select the best arrangement for Fraunhofer's needs and growth.

I am proud to working as the local Mathematica partner with Fraunhofer from the very beginning. It is a continuous dynamic exchange of requirement specification and services.

One Size Does Not Fit All

What data mining and machine learning tasks have in common: extract models from data. But the purpose varies from analytics, over predictive modeling to even control. And there is no way that you find one method which covers everything.

In analytics you want to gain insight. You want understandable, interpretable models, which tell you about the most influencing parameter and detect the pink noise. Rule and tree based methods are quite good for that. But to understand models better, you need to test them. Consequently, they also need to be computational. Fuzzy variants of rule, tree, cluster models are (by the use of fuzzy inference controllers).

For predictive modeling your models need to be computational. If we use fuzzy methods we apply regularization techniques to numerically optimize the fuzzy rules. So, we get accurate predictions by interpretable models. But highly accurate predictors might need, say, neural networks.

To calculate control functions for given goal parameters you need to apply an inverse problems view. To solve the backwards problems by creating many forward paths needs computational power and clever identification techniques.

This is why our machine learning framework comes with a suite of ml methods; kernel methods are numerically optimized in C++. They are integrated into Mathematica's declarative environment to make it easy to solve complex data mining tasks by arrangements of ml methods that are parallelized. Mlf task builders support ml developers to create such arrangements, meta-learning algorithms and automate cross-model testing ...

But even more, to extend the availability of ml methods, we provide a Mathematica-to-WEKA interface.

Understand Your Processes Better

In modern industrial processes, massive data are gathered to monitor and analyse a process. Intelligent analysis helps to reveal as much information of the process as possible. This information is most useful, when it is available in the form of interpretable, predictive models. Machine learning is about extracting such models from data automatically. And our, now, grid-enabled, machine learning framework (mlf) provides a suite of ml methods, which create models that are interpretable AND computational, following the black-box and white-box principles.

Paper production is a complex flow process with many process steps and levels and hundreds of process parameters influencing dozens of quality features. Paper production is so complex that the creation of explicit models is hardly possible. Intelligent data analysis, by means of ml is the right way to overcome this barrier.

For the Technology Center of VOITH Paper, a market-leading provider of large paper production systems, who built a project consortium with SCA, we conducted a "PaperMiner" project.

PaperMiner combines a GUI type fromt-end and an expert front-end. The expert front-end enables the users to configure ml tasks. This is very intuitive through mlf's Mathematica front-end.

The greatest concrete result of the PaperMiner project. SCA set a new generation paper machine, from VOITH, in operation and introduced a new high-quality Graphics paper. Due to better insight, derived from the PaperMiner support, the President of SCA SC Laakirchen could summarize, "This was an unexpected short start-up", his appreciation.

Hidden in Plain Sight II

Innovative business principles? I like the demand-first model described in Hidden in Plain Sight .

From the book's by-text: Every layer of creativity, every success can inch a company farther from its customers.

I really second. Think of the success of Mortage Backed Securities, Credit Derivatives, .. that led to uncontrolled entity leverages and finally the crises which has shifted markets to new regimes characterized by almost unimaginable anomalies.....

From the book and how I applied this to our machine learning framework for Mathematica (mlf) business:

Create a demand landscape through an enhanced understanding of how people behave, their needs, wants, passions and frustrations. Production engineers make products, thus margins. Quality people check products, just before they are rolling out. Consequently, they are seen as the last hurdle. Mlf supports quality people with inline-process quality assurance. They trigger important decisions in early stages.

Reframe the opportunity space by looking with the eye of the customer, the eye of the market and the eye of the industry. Process and quality people collaborate tightly and make better control decisions (eye of the mlf customer). The market recognizes that mlf users deliver uncompromising quality (eye of the customer’s customers). Competition recognizes that mlf customers' in-process quality assurance is a success factor (eye of the process industry).

Formulate a strategic blueprint, make new opportunities relevant in the socio-economic context. We develop on requirements of process sectors (like steel, pulp and paper, energy) and co-finance such individual projects (in the frame of a industrial-research center). The transformation of industry-level experiences keep mlf up-to-date.

Summarising, we do not sell a product, but the transformation of customer experiences, connections and know-how packages.