We are very happy to announce that Eric Postma, Carme Torras and Christian Stöcker will give keynote lectures during the PPSN confernce. Please read more about them and their talk below.
Tilburg University & Jheronimus Academy of Data Science
Parallel Problem Solving in Deep Learning
Artificial neural networks are coarse abstractions of their natural counterparts. The recent deep learning networks at the heart of the AI hype have been shown to be highly effective on “narrow” tasks, i.e., tasks that do not require broad contextual knowledge. Supervised deep learning realises a mapping from inputs onto desired outputs by means of a deep cascade of parameterised nonlinear functions. One of the breakthrough tasks involved the mapping of natural images on appropriate labels describing their main contents. The key difference between deep learning and traditional machine learning models is the number of free parameters used. Whereas traditional machine learning algorithms adhere to Occam’s principle of keeping the number of free parameters as small as possible to avoid overfitting, typical deep learning networks violate this principle by including millions of free parameters. The presentation discusses how this over-parameterisation contributes to solving narrow tasks. The far-reaching implications for interpreting biological and artificial vision systems are explained in terms of the fitness and loss functions involved.
Eric Postma is professor in Artificial Intelligence at the Cognitive Science & AI department at Tilburg University and at the Jheronimus Academy of Data Science in ‘s-Hertogenbosch, a joint initiative of Eindhoven University of Technology and Tilburg University.
His Ph.D. in 1994 at Maastricht University concerned a biologically inspired model of covert attention. This served as an inspiration for his current research, which focusses on the use of data science (machine learning) in image recognition and cognitive modelling. Next to the development of models and theoretical frameworks, he always has and had an open eye for applied research. For instance, starting in the mid 1990s, he focused on the development of visual recognition and classification techniques for the cultural heritage. Subsequently, he launched the development of digital analysis methods for paintings. As a direct consequence, he was the leader of several scientific NWO projects in the programme ToKeN (“Toegankelijkheid en Kennisontsluiting in Nederland”) and CATCH (Continuous Access to Cultural Heritage). Moreover, together with prof. C. Richard Johnson Jr., he initiated an international consortium for digital painting analysis. The work by his research group has been covered extensively in the media since 2000. In 2008 Postma and his team ranked second in the annual “Academische Jaarprijs” with a presentation on the breakthroughs of digital painting analysis. Together with Laurens van der Maaten, he received the AAAI-08 Most Innovative Video Award for a scientific video of the digital painting analysis. Currently, Postma is coordinating the REVIGO project and he is member of SIGAI, IPN, and the Lorenz center Computational Science Board.
Institut de Robòtica i Informàtica Industrial (CSIC-UPC)
Assistive Robotics: AI Challenges and Ethics Education Initiatives
Robotics research has evolved a lot in the last 20 years, and the focus has shifted from the mechanics, kinematics and control of industrial manipulators to the capacities of perception, learning and interaction with people of so-called social robots. This poses a series of research challenges. The way to instruct these robots must be easy and intuitive so that non-expert users can teach them the tasks they have to do, for example, through demonstrations. Not being caged like their predecessors in factories, they must be intrinsically safe to people, a critical aspect of considerable technical difficulty, especially when the interaction requires physical contact. They should be able to perceive and manipulate the deformable objects that abound in domestic and healthcare environments, which involves great complexity given the infinite dimensionality of their shape spaces compared to the six degrees of freedom that characterize the pose of a rigid object. They must also be tolerant to noisy perceptions and inaccurate actions and be endowed with a strong learning ability and adaptability to dynamic environments. Finally, the behaviour of these robots should not be limited to a fixed sequence of actions, but must be goal-driven and collaborative with people.
These six challenges can be pairwise grouped and translated into three U-turns of research in Artificial Intelligence. 1) Usability turn: from exhaustive programming taking into account all situations and rigid control schemes, to learning from demonstrations and compliant control. 2) Uncertainty turn: from high-resolution perception and accurate manipulation planning to task-oriented perception and probabilistic re-planning adapted to context and user. 3) Understanding turn: from building associations to attaching semantics to perceptions of objects and situations as well as reasoning about functionalities and goals.
In the keynote, these U-turns will be illustrated through results from the project CLOTHILDE , which deals with the robotic handling of garments in the healthcare and logistics contexts, an application that requires addressing the six technical challenges mentioned above.
This shift from industrial to social robotics poses also ethical and social defies, which have led to a necessary confluence with the humanities. In addition to establishing regulations and standards, numerous educational initiatives have emerged, where science fiction often plays a prominent role by highlighting the pros and cons of possible future scenarios. In the context of university education, MIT Press has recently published my novel , along with ancillary materials to teach a course on “Ethics in Social Robotics and Artificial Intelligence”. The goal is to provide useful guidelines for students and professionals (robot designers, manufacturers, and programmers), as well as for end users and the general public.
 CLOTHILDE Project (2018-22): https://clothilde.iri.upc.edu/
 C. Torras: The Vestigial Heart. A Novel of the Robot Age. MIT Press, 2018. http://mitpress.mit.edu/books/vestigial-heart
Carme Torras is Research Professor at the Institut de Robòtica i Informàtica Industrial (CSIC-UPC) in Barcelona, where she leads a research group on assistive and collaborative robotics. She received M.Sc. degrees in Mathematics and Computer Science from the University of Barcelona and the University of Massachusetts, respectively, and a Ph.D. degree in Computer Science from the Technical University of Catalonia (UPC). Prof. Torras has published six research books and about three hundred papers in robotics, machine learning, geometric reasoning, and neurocomputing. She has supervised 19 PhD theses and led 16 European projects, the latest being her ERC Advanced Grant project CLOTHILDE – Cloth manipulation learning from demonstrations. Prof. Torras is IEEE and EurAI Fellow, member of Academia Europaea and the Royal Academy of Sciences and Arts of Barcelona. She has served as Senior Editor of the IEEE Transactions on Robotics, and has played different roles in the editorial boards of 10 journals, among which AI Communications, Robotics and Autonomous Systems, and Natural Computing. Convinced that science fiction can help promote ethics in AI and robotics, one of her novels – winner of the Pedrolo and Ictineu awards – has been translated into English with the title The Vestigial Heart (MIT Press, 2018) and published together with online materials to teach a course on “Ethics in Social Robotics and AI”.
Professor of Digital Communication
Competence Center Communication (CCCOM)
AI and Academia – are you ready to become toolmakers?
The machine learning landscape has changed and diversified in unexpected ways in the last few years. As machine learning methods become viable tools for other researchers, for areas ranging from material science to genomics, the demand for application oriented AI researchers is growing dramatically. At the same time, large companies offer fantastic salaries and working conditions for people who can credibly claim to be able to develop machine learning solutions for specific problems. Where do you go from here? Pure research, making tools, making money? A brief look at the most exciting field in informatics – from the outside.
Christian Stöcker is Professor of Digital Communication at the Competence Center Communication (CCCOM) at Hamburg University of Applied Sciences. There he is establishing a newsroom and in charge of a new master program. Besides, he is a columnist at Spiegel Online (one of the most widely read German-language news Websites) where he also had been Head of the Internet Department from 2011 to 2016. Christian Stöcker has published several books within the context of digitization’s impact on society and is one of few German journalists who had access to Edward Snowden’s archive of intelligence documents. He co-authored several investigative stories about the NSA, GCHQ, and other intelligence services.