Home > Program > Plenary Talks
Plenary Talks
[ Prof. Heinrich Bülthoff, Max Planck Institute for Biological Cybernetics, Germany ]
Title: Towards artificial systems: what can we learn from human perception
The question of how we perceive and interact with the world around us has been at the heart of cognitive and neuroscience research for the last decades. Despite tremendous advances in the field of computational vision made possible by the development of powerful learning techniques as well as the existence of large amounts of labeled training data for harvesting - artificial systems have yet to reach human performance levels and generalization capabilities. In this contribution we want to highlight some recent results from perceptual studies that could help to bring artificial systems a few steps closer to this grand goal. In particular, we focus on the issue of spatio-temporal object representations (dynamic faces), face synthesis, as well as the need for taking into account multi-sensory data in models of object categorization. Having understood the important role of haptic feedback for human perception, we also explored new ways of exploiting it for helping humans (pilots) in solving difficult control tasks. This recent work on human machine interfaces naturally extends to the case of autonomous or intelligent machines such as robots that are currently envisioned to be pervasive in our society and closely cooperate with humans in their tasks. In all of these perceptual research lines, the underlying research philosophy was to combine the latest tools in computer vision, computer graphics, and virtual reality technology in order to gain a deeper understanding of biological information processing. Conversely, we discuss how the perceptual results can feed back into the design of better and more efficient tools for artificial systems.

Heinrich Bülthoff is scientific member of the Max Planck Society and director at the Max Planck Institute for Biological Cybernetics in Tübingen. He is head of the Psychophysics Department in which a group of about 70 biologists, computer scientists, mathematicians, physicists and psychologists work on psychophysical and computational aspects of perception and cognition. Prof. Bülthoff is involved in many international collaborations and member of European research networks and many projects funded by the European Commission.
He holds a Ph.D. degree in the natural sciences from the Eberhard-Karls- Universität in Tübingen. From 1980 to 1988 he worked as a research scientist at the Max Planck Institute for Biological Cybernetics and the Massachusetts Institute of Technology. He was Assistant, Associate and Full Professor of Cognitive Science at Brown University in Providence from 1988-1993 before becoming director at the Max Planck Institute for Biological Cybernetics. Since 1996 he is also Honorary Professor at the Eberhard-Karls- Universität in Tübingen and from 2004 to 2009 Editor in Chief of the ACM Transactions on Applied Perception. Since 2009 he is also Distinguished Visiting Professor at the Department of Brain and Cognitive Engineering at Korea University.

[ Prof. Mitsuru Ishizuka, Univ. of Tokyo, Japan ]   http://www.miv.t.u-tokyo.ac.jp/ishizuka/eng.html
Title: Exploiting Macro and Micro Relations toward Web Intelligence
Relations are basic elements for representing knowledge, such as in semantic network, logic and others. In Web intelligence research, the extraction or mining of meaningful knowledge and the utilization of the knowledge for intelligent services are key issues. In this talk, I will present some of our researches related to these issues, ranging from macro relations to micro ones. Here we mostly use Web texts, and the use of their huge data though a search engine becomes a key function as together with text analysis. The first topic concerns with the extraction of human-human and company-company relations from the Web. Relation types between two entities are also extracted here. An open Web service based on this function has been operated in Japan by a company. One technology related to this one is namesake disambiguation. Wikipedia is a good reliable source for wide knowledge, unlike other Web information. In order to extract the knowledge or data from Wikipedia in the form that computers can understand and manipulate, several attempts including ours have been carried out, typically to extract triplets such as (entity, attribute, value). After we worked on computing similarity between two words based on the distributional hypothesis, we have been interested in computing similarity between two word pairs (or two entity pairs). Like in the previous case, we are mainly utilizing distributional hypothesis, and have invented an efficient clustering method for dealing with several tens of thousands of lexical patterns. Based on this mechanism, we have implemented a latent relational search engine, which accepts two entity pairs with one missing component such as {(Tokyo, Japan), (?, France)} as a query, and produces an answer such as (? = Paris) with its evidence. As an extension of this mechanism, we recently invented an efficient co-clustering method, which works well to find arbitrary existing relations between two nouns in the sentences. This problem setting is called open information extraction (open IE). The final topic of the talk is Concept Description Language (CDL), which has been designed to serve as a common language for representing concept meaning expressed in natural language texts. Unlike Semantic Web which provides machine-readable meta-data in the form of RDF, CDL aims to encode the meaning of the whole texts in a machine-understandable form. The basic representation element in CDL is micro relations existing between entities in the text; 44 relation types are defined. CDL has been discussed in a W3C incubator group for international standardization since 2007. It is intended to be a basis of semantic computing in next generation, and also become a medium for overcoming language barrier in the world. Current issues of CDL are, among others, an easy semi-automatic way of converting natural language texts into the CDL description, and an effective mechanism of semantic retrieval on the CDL database.

Mitsuru Ishizuka is a professor and the vice-dean of the School of Information Science and Technology, the Univ. of Tokyo. Prior to this position, he was at NTT Yokosuka Lab. and the Institute of Industrial Science, the Univ. of Tokyo. His research interest is in artificial intelligence in general; previously he worked on hypothetical reasoning, and presently is working on Web intelligence and intelligent multimodal media with lifelike agents. He earned his B.S., M.S. and Ph.D. from the Univ. of Tokyo. He is a past president of JSAI (Japanese Society for Artificial Intelligence), and was the program co-chair of PRICAI-2002.

[ Prof. Toby Walsh, NICTA, Australia ]  http://www.cse.unsw.edu.au/~tw/
Title: Symmetry
Symmetry can be used to help solve many problems. For instance, Einstein's famous 1905 paper ("On the Electrodynamics of Moving Bodies") uses symmetry to help derive the laws of special relativity. In artificial intelligence, symmetry has played an important role in both problem representation and reasoning. In this talk, I will describe recent work on using symmetry to help solve combinatorial optimization problems. I will discuss both symmetries within individual solutions of problems as well as symmetries between different solutions of the same problem. I will show how such symmetry reasoning can speed up reasoning, and has led to the discovery of new results in both graph and number theory. Some of the work described is joint with Marijn Heule, George Katsirelos and Nina Naroditskaya.

Toby Walsh is currently Scientific Director of NICTA, adjunct Professor at the University of New South Wales, external Professor at Uppsala University and an honorary fellow of Edinburgh University. He has been Editor-in-Chief of the Journal of Artificial Intelligence Research, and AI Communications. He has been elected a fellow of both the Association for the Advancement of Artificial Intelligence and the European Coordinating Committee for AI. He has been Secretary of the Associtation for Constraint Programming (ACP) and is currently Editor of its Newsletter. He is one of the Editors of the Handbook for Constraint Programming, and the Handbook for Satisfiability. He has been Conference and/or Program Chair of the CP, SAT and IJCAR conferences. He will be Program Chair of IJCAI 2011.

[ Dr. Mike Schuster, Google, USA ]  
Title: Speech and Research at Google
In this talk first I will talk about some of the speech-related projects that are going on in Google Research, some of them already publicly launched. After a mini-introduction into speech recognition I will talk about GOOG-411, a phone-based directory assistance service launched in the US and Canada, and VoiceSearch, a recently launched service as part of the Google Mobile App, which allows free-from input search on the iPhone/iPod Touch or Android. I will show demos, some videos, and talk about some before/after launch challenges. I will also briefly introduce some audio indexing experiments and launches that have been done for YouTube and transcription of voice mail as part of the Google Voice Project. Finally I will talk about Google's general approach to research and how speech and other focus areas fit into it.

Mike Schuster graduated in Electric Engineering from the Gerhard-Mercator University in Duisburg, Germany in 1993. After receiving a scholarship to study in Japan he spent a year in Japan to study Japanese in Kyoto and Fiber Optics in the Kikuchi laboratory at Tokyo University. His professional career in speech brought him to Advanced Telecommunications Research Laboratories in Kyoto, Nuance in the US and NTT in Japan where he worked on speech recognition research and development after getting his PhD at the Nara Institute of Science and Technology. Today he is part of the speech recognition group at Google Research where he focuses on all aspects of making speech recognition technology available to the public, recently mostly through the VoiceSearch project.