Plenary Lectures

 Scene variability and perception constancy in the visual system: a model of pre-processing before data analysis and learning
 Cognitive components of digital media
 Kernel machines and their applications

Scene variability and perception constancy in the visual system: a model of pre-processing before data analysis and learning

Jeanny Herault Professor Jeanny Herault
University Joseph Fourier, Grenoble, France
Personal homepage


Jeanny HERAULT is professor emeritus of the University Joseph Fourier, Grenoble. He taught signal processing in engineering schools and presently gives a series of lectures in visual perception at the Master's degree of Cognitive Science. Since 1968, his field of research has been concerned with modeling of Natural and Artificial Neural Networks, at various levels: from cellular membrane to large adaptive networks. His interests range from theoretical studies, computer simulations, design and implementation of electronic analog and digital neural machines, to applications in Image and Signal Processing. Since early 90's he has been interested in models of visual perception (retinal processing, motion estimation, color and cortical processing of natural scenes) as well as in high-dimensional data processing through self-organizing neural networks. He is member of several scientific committees, expert of the European Community for Future and Emerging Technologies projects and Scientific Advisor for industrial companies. He is reviewer for international journals in Neural Networks, Signal and Image Processing.

Abstract

Hell in data analysis is paved (at least) with variability and noise. Is there some lost garden of Eden? Is there some way to approach it? In this talk, I take the example of Human visual perception and I show how our visual system manages to process visual information in such a highly efficient way that it is able to categorize images or scenes within ranges of 100-150 ms, whatever the viewing conditions. In fact, before any high-level recognition task, the visual system unfolds a series of preprocessing stages to reduce image variability:

We can consider images as represented by high-dimensional vectors, the components of which are the image's energy extracted by each frequency filter. Taking advantage of all these preprocessing stages, data variability can be expected to be optimally reduced for comparison purposes. In the last part of this talk, I give an example of image categorization by means of CCA, a self-organizing neural network, which, after noise reduction, reduces dimension and unfolds the manifold where the data are embedded. It is shown that the network, taught by human subjects performing the same task, is able to improve its behavior by providing a better separation of classes.

Cognitive components of digital media

Lars Kai Hansen Professor Lars Kai Hansen
Technical University of Denmark, Denmark
Personal homepage


Professor Lars Kai Hansen received the PhD degree in physics from University of Copenhagen, in 1986. He worked on industrial machine learning from 1987-1990, with Andrex Radiation Products A/S. Since 1990 he has been with the Technical University of Denmark, currently he is head of DTU Informatic's Section for Intelligent Signal Processing. Lars Kai Hansen is author/co-author of more than 200 papers and book chapters on adaptive signal processing and machine learning and applications in bio-medicine and digital media.

Abstract

Among the exponentially many ways of grouping data, can we characterize the ways that are likely to make sense to a human?. This is a classical research question in psychology -- going back at least to the Gestalt Theorist. Cognitive component analysis is a quantitative research program in which we apply unsupervised learning methods to digital media to understand the conditions under which learned structure is well-aligned with human cognitive activity. In the talk I will introduce machine learning methods for cognitive component analysis and present evidence for cognitive components in abstract data such as text, social interactions, music, and speech. I wil demonstrate a number of applications in human computer interfaces including specialized search engines for music, spoken documents, and neuroimaging data.

Kernel machines and their applications

Klaus-Robert Mueller Professor Klaus-Robert Müller
Technical University Berlin, Germany
Personal homepage


Klaus-Robert Müller has been Professor for Computer Science at TU Berlin since 2006; at the same time he is directing the Bernstein Focus on Neurotechnology Berlin. He studied physics in Karlsruhe from 1984-89 and also obtained his PhD in Computer Science at TU Karlsruhe in 1992. After a PostDoc at GMD FIRST in Berlin from 1992-1994, he was a European Community STP Research Fellow at University of Tokyo from 1994-1995. From 1995 he built up the Intelligent Data Analysis (IDA) group at GMD FIRST (later Fraunhofer FIRST) and and directed it until 2008. 1999-2006 he was a Professor for Computer Science at University of Potsdam. In 1999, he was awarded the Olympus Prize by the German Pattern Recognition Society, DAGM and in 2006 he received the SEL Alcatel Communication Award. He has co-authored more than 250 peer reviewed papers and is active in numerous program committees and editorial boards. His research interests are intelligent data analysis, machine learning, statistical signal processing and statistical learning theory with the application foci computational finance, computational chemistry, computational neuroscience and genomic data analysis. Since 2000 one of his main scientific interests is to study the interface between brain and machine: non-invasive EEG-based Brain Computer Interfacing.

Abstract

This lecture provides a very brief introduction to Support Vector Machines as an example for successful kernel-based machine learning (ML) and touches on fundamentally open issues in this field.

Then I review in short selected successful applications of kernel based ML, i.e.for hacker intrusion detection, computational chemistry etc.

The main part of my talk will discuss ML methods for the online analysis of brain signals, i.e. I chart the path towards an EEG-based Brain Computer Interface.

Brain Computer Interfacing (BCI) aims at making use of brain signals for e.g. the control of objects, spelling, gaming and so on. In particular the talk will show the wealth, the complexity and the difficulties of the data available, a truely enormous challenge: In real-time a multi-variate very strongly noise contaminated data stream is to be processed and neuroelectric activities are to be accurately differentiated.

Finally, I report in more detail about the Berlin Brain Computer (BBCI) Interface that is based on EEG signals and take the audience all the way from the measured signal, the preprocessing and filtering, the classification to the respective application. BCI as a fascinating new channel for man-machine communication is discussed in a clincial setting and for human machine interaction.


 Download handout -->