1. ENGLISH
  2. News
  3. Stories
  4. 2018
  5. 11
  6. Learning to learn

We use cookies (our own and those of third parties) to make our websites easier for you to use and to display advertisements in accordance with your browser settings. By continuing to use our websites, you consent to the use of cookies. Please see our Cookie Policy for more information on cookies and information on how you can change your browser's cookie settings: Cookie Policy Accept

Learning to learn

Volkswagen AG opened the highly specialized research center known as Data:Lab Munich around five years ago with a focus on data analytics. The center’s team of experts is now working on developing artificial intelligence.

Data:Lab Munich develops artificial intelligence

Professor Patrick van der Smagt can break down complex topics in a way that lay people can understand. That is a useful skill to have when explaining the subject of artificial intelligence and machine learning. “Do you know why computers can tell on their own these days whether you’ve posted a photo of a cat or a dog on Facebook?” asks this Dutch researcher and then promptly provides the answer. “Because people comment on it in their posts, and systems can learn from that – faster and better than any user.” By the way, that same principle applies to stock market forecasts, CT scan analyses, and language recognition programs.

Teaching systems to analyze comments

Professor Patrick van der Smagt left a university research position to join Volkswagen’s Data:Lab Munich in 2016

Algorithms are only powerful if they are properly fed. Precisely that is the challenge for the artificial intelligence (AI) research team led by van der Smagt at Volkswagen’s Data:Lab Munich. “In contrast to social networks, if you’re trying to do something like quality control for automotive components you don’t have a stock of what’s called annotated data. So we have to teach the systems to analyze comments, determine contexts, and draw the best possible conclusions on their own.”

The algorithms process information in a series of steps and combine it with what they have already learned plus historical data sets. In so doing, the systems become increasingly good at identifying patterns and regularities.

Data:Lab Munich was founded in late 2013 – at a time when artificial intelligence had a much lower public profile than it does today. The center initially focused on data analytics, and its experts were still analyzing complex data themselves. The team has been growing ever since. Today its approximately 80 IT specialists, data scientists, programmers, physicists, and mathematicians are pursuing research and developing applications in the field of machine learning.

This field is of interest to Volkswagen because artificial intelligence is becoming ever more relevant to industry. Already a competitive factor, it will only gain in significance. Volkswagen has therefore laid the foundations at an early stage for building and applying its own high-performance AI systems.

“We assess every single AI-based application to see how helpful it can be,” says Martin Hofmann, CIO of the Volkswagen Group. “Our aim is to use AI to provide even better support for our teams at factories and offices. Because that will generate lasting benefits for the entire company.”

Algorithms forecast sales for different models and regions

Firas Lethaus heads the team working on industrial applications for machine learning

While van der Smagt is responsible for basic research at Data:Lab Munich, Firas Lethaus is in charge of transferring the technology to concrete areas of application. “We’re the anchor point for the field of artificial intelligence at the Volkswagen Group,” explains Lethaus, who is the head of Deep & Machine Learning. “All the brands and departments in our company can make use of our expertise. We also turn around and approach them with our ideas on where and how machine learning can be applied to their operations. The idea behind all of this is to give them the necessary tools to make faster and better decisions.”

Concrete applications are already being used in areas like sales planning. Algorithms can predict which car models will sell especially well in certain seasons in particular regions. Their calculations incorporate context-based data – such as growth forecasts, economic sanctions, and weather conditions. The results give employees a substantial basis of data with which to make strategic decisions.

Another example is the aftersales business. Which replacement parts are in especially high demand where and when? Lethaus describes the logic behind machine learning as follows: “In hilly regions cars have greater wear on their brakes. The system concludes from this that the warehouses need to carry more brake discs. Or take the use of horns. In Europe drivers don’t use their horns very often, but in China they do all the time – so there’s high wear on those components and more demand for replacements.” Data:Lab teams have now developed more than 100 AI applications, many of which are going through trial runs in very different parts of the company.

Journey into the future

The Data:Lab’s multidisciplinary team includes not only IT experts but also mathematicians, physicists, and data scientists

Another important application for AI lies in processing the camera images needed in fields such as quality assurance or autonomous driving. “It doesn’t really matter to a neural network what it sees. It derives contexts on the basis of pixel data. The challenge consists of structuring the data in such a way that they are sufficiently robust to prompt the right actions in specific situations. If a car camera sees that a cyclist is crossing the street, the system can’t tell from the pixels where the cyclist will be in the next five seconds. That’s much too complex. But the dynamics of the movement can be abstracted from the image,” explains van der Smagt.

And where is the Data:Lab’s fascinating journey expected to lead? According to Lethaus we’re just setting off on a journey into the future. “What are we after? Self-learning systems that offer people the best possible basis for their decisions. We’re on the right course, but there’s a long way to go before we arrive.”
 

Artificial intelligence is relevant to the Volkswagen Group in two major areas:

  • in the vehicle and automotive technology
  • in business processes and applications in the company
  • Artificial intelligence in the vehicle:

    AI is central to autonomous driving, i.e. the use of self-learning algorithms for environmental detection and control of the vehicle. A further application area in the vehicle is Natural Language Processing (NLP) - advanced, intelligent voice control, for example for navigation and comfort functions. The use of artificial intelligence is also conceivable for face or voice recognition in order to unlock or start the vehicle.

  • Artificial intelligence in the company:

    “Cognitive ergonomics" is the term Volkswagen uses to describe the use of self-learning algorithms that relieve employees of small, repetitive administrative tasks and prepare them for the final decision (form releases, entries, etc.). These programs are called bots.

    The use of AI is also useful for information processing: algorithms can search and organize highly complex data sets quickly, efficiently and continuously. This enables, for example, precise market forecasts containing a multitude of variables (economic development, household income, customer preferences, model availability, price level, etc.). Artificial intelligence can thus prepare a substantial database for the human decision maker and present it for decision.

    In addition, the use of artificial intelligence in manufacturing is an important area of application for Volkswagen as a large industrial company. This involves, for example, intelligent robotics, i.e. artificial intelligence that enables robots to work hand-in-hand with human skilled workers. It is also about powerful algorithms that can optimize manufacturing and logistics processes even more efficiently than before. Another example is so-called predictive maintenance: tools and machines will be able to learn their optimum maintenance intervals themselves using self-learning algorithms and signal them to the maintenance personnel in the factory.

    Apart from work in the office and factory, artificial intelligence also offers new possibilities in the area of cyber security.