WORKSHOPS

The ICANN 2007 workshops will consist on a panel of invited experts, which after presenting the subject will moderate an open debate.

WORKSHOP 1

Cognitive Systems
John G. Taylor, King's College, London, john.g.taylor(a.t)kcl.ac.uk

Background
Cognitive systems deserve a lot of attention (and emotion): they require higher-order control (as through attention) on top of lower level perception and concept forming, as well as value-biasing (as through emotional values of stimuli) so as to achieve task solutions of value to the cognitive system. At the same time there is the question as to embodiment or not? This is a difficult challenge: without embodiment a system is powerless to build up a suitable repertoire of responses by which it can reason/think of how to solve problems in its environment. At least some sort of movement is desirable, although may be limiting in terms of present-day robotic technology, both mechanically and computationally. There are presently a large number of EC Projects supported to help the creation of Cognitive Machines, as part of the Cognitive Systems Unit effort. All of these are potential contributors, especially those attacking the problem of cognition from a neural network point of view.

Goals
This leads us to tackle a set of basic questions about cognitive systems that need to begun to be answered through computational experience:

1) What are the crucial components of cognition?
2) Must a cognitive system be embodied to be termed 'cognitive'?
3) How much should we take from brain science in implementing a cognitive system?
4) Is language important for the development of cognition in a machine, and how might language be implemented?
5) How important is emotion in the development of a cognitive system able to interact in an 'ethical' manner with those around it, and how might emotions be implemented?
6) How can machine cognition be achieved in terms of robotic an implementation?
7) What is the basis of a reasoning system, and how would it be implemented?
8) Is consciousness necessary for a truly cognitive system, as would appear in humans)?
9) What are the ethical problems that should be resolved in creating cognitive machine systems?

The proposed Workshop on Cognitive Systems will take the same successful format as before: a set of prepared position papers by a panel on some or all of the questions raised at the ICANN06 Workshop (similar to those listed above), and then a discussion on the consensus of possible answers to the various questions from any who wished to contribute from the floor or the panel, as well as other questions of relevance to the topic.

Target Audience
There are presently a large number of EC Projects supported to help the creation of Cognitive Machines, as part of the Cognitive Systems Unit effort. All of these are potential contributors, especially those attacking the problem of cognition from a neural network point of view. Other groups interested in this area will also be targeted, such as those in Vision Research, as well as the Artificial Intelligence community associated with the First Artificial Intelligence Conference (AG1-08) and related colleagues across the world.

WORKSHOP 2

Neural Networks in Biomedical Engineering and BioInformatics
Sepp Hochreiter, Johannes Kepler University, Linz, hochreit(a.t)bioinf.jku.at
Paulo Cortez, University of Minho, Portugal, pcortez(a.t)dsi.uminho.pt

Background
Bioinformatics is a relative new, interdisciplinary research area which bridges the gap between life sciences and computer science. It was a key technology for one of the most important achievements of mankind, the decoding of the human genome.
For genome sequencing bioinformatics focused on algorithmic aspects, however these days its main challenge is to analyze data from life science, medicine, and pharmacy industry.
To analyze these data, neural networks are one of the essential bioinformatics tools, e.g. for secondary and 3D structure prediction of proteins. But modern measurement techniques in both biology and medicine like microarrays, SNP (single nucleotide polymorphisms) data or mass spectrometry data require new approaches to data analysis, that is they ask for new neural network and machine learning techniques. For example new methods for feature selection must be developed both for building biological models and for prediction.

Goals and Target Audience
The goal of this workshop is to bring members of the neural network community which are interested in medical applications and bioinformatics together in order to reflect and discuss about current challenges and open problems -- and how to solve them.

WORKSHOP 3

What it means to communicate
Stefan Wermter, University of Sunderland, Stefan.Wermter(a.t)Sunderland.ac.uk - (Contact for information)
Vittorio Gallese, University of Parma, Vittorio.Gallese(a.t)ipruniv.cce.unipr.it
Friedemann Pulvermuller, MRC, Cambridge, Friedemann.Pulvermuller(a.t)mrc-cbu.cam.ac.uk
Mark Elshaw, University of Sunderland, Mark.Elshaw(a.t)Sunderland.ac.uk
Christo Panchev, University of Sunderland, Christo.Panchev(a.t)Sunderland.ac.uk
Mike Knowles, University of Sunderland, Michael.Knowles(a.t)Sunderland.ac.uk
Martin Page, University of Sunderland, Martin.Page(a.t)Sunderland.ac.uk

WEB: http://www.his.sunderland.ac.uk/nestcom/workshop/

Background
This workshop will focus on the neural, computational and cognitive principles of communication. A wealth of new knowledge has been produced by recent research for instance as part of the EU NEST initiative "What it means to be human". The goal of this workshop is to explore relevant research to focus on the question "what it means to communicate". The general aim is to understand the neural, cognitive, social, computational and developmental features that have led to communication differences between humans and animals. A number of interesting and successful research directions have been supported including learning by imitation, examining the origin of human rule based reasoning, studying the neural origins of language, exploring the evolutionary origins of the human mind, researching into verbal and nonverbal communication, using and interpreting signs, characterising human language by structural complexity, and representing abstract concepts. To complement this, the NESTCOM project aims to understand the results of these projects and integrate them with a neural multimodal understanding of verbal and visual communication for embodied action understanding.

Goals
This workshop will explore the characteristics of human communication and their relationship to the role of networks of mirror neurons.
Topics of discussion could include:

1) Neural or cognitive language modelling
2) Neural or cognitive vision modelling
3) Multimodal integration
4) Bioinspired communicating robotics
5) Developmental approaches to communication
6) Imitation and Learning
7) Signs and Gestures for communication
8) Mirror neuron system and its role in multimodal integration
9) Learning mechanisms for communications skills
10) Learning of action understanding

Target Audience
This workshop is aimed at researchers with an interest in any aspect of Neural Networks and Communication, including Multimodal Communication, Speech, Gestures etc. We encourage researchers from a variety of fields working in Neuroscience, Psychology, Intelligent Systems and Computing or any other relevant sphere.

WORKSHOP 4

Neural networks of the future?
Wlodziszlaw Duch, Torun PL & Singapore, wduch(a.t)is.umk.pl
Erkki Oja, Helsinki University, FL, Erkki.Oja(a.t)hut.fi

Background
Neural networks have gone a long way in the last half a century. It is important to have clear direction and therefore there is a need for a panel discussing perspectives of the field.
Neural networks have branched into many areas, giving inspirations to machine learning (kernel systems, signal analysis) on one extreme, and to computational cognitive neuroscience with rather faithful biophysical models on the other side. In recent years most of the improvements of current models are still concerned with learning in relatively simple situations, from single dataset. Although we can do classification and approximation quite well with many learning models we still do not have systems that can learn from natural perception, or can learn difficult logical problems. One direction is to look for neurocognitive inspirations at higher level than single neurons. What are the most promising directions to reach human-level competence?
Important advances in recent years have been made in biologically inspired vision, auditory scene analysis, neural hardware and many other directions.
Where should we concentrate our efforts?

Goals and Target Audience
We invite all participants to send us questions they would like to discuss and invite top experts to answer these questions. We do not plan position statements to have more time for real discussion. We shall post these questions at this web page, encouraging participants and panelists to add more questions and comments.

Print Page