Brain Inspired Cognitive Systems 2004
29 August - 1 September 2004, University of Stirling, Scotland, UK

BICS 2004: Plenaries

Fifteen years of Neuromorphic Engineering: progress, problems, and prospects

Slides not available

Prof Rodney Douglas, Institute of Neuroinformatics, University/ETH, Zurich, Switzerland

Fifteen years of Neuromorphic Engineering: progress, problems, and prospects Neuromorphic engineers design and fabricate artificial neural systems: from adaptive single chip sensors, through reflexive sensori-motor systems, to behaving mobile robots. Typically, knowledge of biological architecture and principles of operation is used to construct a physical emulation of the target neuronal system in an electronic medium such as CMOS analog very large scale integrated (aVLSI) technology. Initial successes of neuromorphic engineering have included smart sensors for vision and audition; circuits for non-linear adaptive control; non-volatile analog memory; circuits that provide rapid solutions of constraint-satisfaction problems such as coherent motion and stereo-correspondence; and methods for asynchronous event-based communication between analog computational nodes distributed across multiple chips. These working chips and systems have provided insights into the general principles by which large arrays of imprecise processing elements could cooperate to provide robust real-time computation of sophisticated problems. However, progress is retarded by the small size of the development community, a lack of appropriate high-level configuration languages, and a lack of practical concepts of neuronal computation.

Attention and Consciousness as Control System Components in the Brain

Prof JG Taylor, Department of Mathematics, King's College, Strand, London WC2R2LS, UK

Slides

My thesis is that by applying engineering control theory to attention in sufficient detail, it will guide us as to how consciousness could be created. I will start by describing experimental data on attention control (single cell, fMRI, etc), and then develop a flexible engineering control model of attention. This will be extended from control of sensory attention to attention to motor response. Simulations will be presented of the overall architecture. Experiments that show that attention is an important gateway to consciousness will then be presented. The attention control model will be extended to the CODAM model, which will be explored as a suitable substrate for conscious experience. This model is then used to explore various features of experiential change in mental diseases (schizophrenia, PD, AD, autism) and how the essential corollary discharge feature on CODAM allows an understanding of modifications of ownership as well as agency in self. The CODAM model leads to a very important function to consciousness: that of speeding up attention. Links will be made to meditation  and mystical experience, and various philosophical problems of mind will be considered in terms of CODAM, as well as implications for animal consciousness.

Computation, cognition, and control

Owen Holland Senior Lecturer Department of Computer Science University of Essex

Slides

We all accept that this discipline is about cognition, and about computation. However, it is all too easy to overlook the central fact that the structure that gave rise to cognitive neuroscience, the human brain, is not primarily a computer or a cognizer, but a control system – a rather unusual control system, in that it was evolved rather than designed. Have we yet taken sufficient account of the true nature and function of the brain? This talk will review current efforts and achievements within the field, from computationally inspired models to brain-inspired computation, and will show how the concepts and constraints associated with a control-centred viewpoint can act both to unify findings and to guide future progress.

Neural Nets: the hype and the reality, from an industrial perspective

Prof. Graham Hesketh Team Leader - Information Engineering Strategic Research Centre Rolls-Royce plc Derby DE24 8BJ United Kingdom

Slides (large file)

Neural networks have had a chequered history. After the early promise of Rosenblatt's perceptrons and the crushing revelations of Minsky and Papert, neural networks had a resurgence in the 80's and 90's with a plethora of usable architectures which threatened to revolutionise our businesses. But the transformation from hype to reality has not been smooth. The path has been littered with triumphs and disasters, and the lessons have been hard. But today, estimates indicate that over 80% of Fortune 500 companies have neural network R&D programmes. This talk gives a personal view of the evolution of neural networks, drawn from 15 years experience of applying them in industry.

Self-Organisation in the Nervous System: the Establishment of Nerve Connections by an Inductive Mechanism

David Willshaw MRC External Scientific Staff & School of Informatics, University of Edinburgh, UK.

No slides available

One of the great unsolved problems of neurobiology is how, during development, the axons from individual nerve cells are able to find their appropriate target nerve cell to form, in many cases, highly ordered patterns of connections. One very powerful idea is that connections are made on the basis of chemical markers carried by the participating cells, which are used for cell-cell recognition.
The recent discovery of Eph receptors, distributed in a graded fashion >across the vertebrate retina, and their associated ligands, the ephrins, which are similarly distributed across the optic tectum or superior colliculus, provides new constraints for contemporary models for the formation of nerve connections. Identification of the possible molecular basis for the formation of nerve connections enables a link to be made between the phenomena (ie, the neuroanatomy of the patterns of connections formed) and the underlying molecular substrate.
In this talk I show that the model for the formation of ordered retinotectal nerve connections by means of the induction of molecular markers from retina onto tectum (developed originally by myself and Christoph von der Malsburg) can be applied successfully to the modern neurobiological findings.
I will demonstrate:

Disentangling signals blindly from nonlinear mixtures

Professor Erkki Oja, Helsinki University of Technology

No slides available

The problem of blind signal separation means that we have a set of parallel signals - temporal or spatial - and we try to find out underlying unknown sources that have been mixed to produce our observations. Usually, the mixing is assumed to be linear, and then the problem can be solved under certain conditions. This is the technique known as Independent Component Analysis. If the mixing is nonlinear, however, then the problem is much harder, in fact intrinsically ill-posed. Yet, there are some approximative techniques which manage to give a reasonable solution even in this case. The problem and some solutions are outlined in the talk, with examples illustrating nonlinear separation.

Mechanical Bodies; mythical minds: dancing with pixies

Dr Mark Bishop, Reader in Computing, Goldsmiths College, University of London

Slides


Back to BICS 2004 page.
Back to ICSC Home Page