The EPFL-LATSIS Symposium 2006 - Dynamical principles for neuroscience and intelligent biomimetic devices

The goal of the conference is to bring together scientists and engineers interested in understanding the dynamical properties of the nervous system, and in taking inspiration from those properties for the design of prosthetic and robotic devices. The conference is interdisciplinary in nature, and aims at bringing together researchers working on similar topics and phenomena but from different backgrounds.

Background

The symposium is interested in understanding the dynamical principles in a variety of topics ranging from neurophysiology, neural computation, neuroprosthetics, hybrid circuits and electronic neurons, to biomimetic robotics and control.

Currently, all these topics are at the forefront of brain-machine research and are being developed more or less independently of one another. However it is obvious that there are many common problems that are related to the dynamics of biological and electronic subsystems. In particular the following are important: (i) the understanding of the dynamics underlying the predictability of motor activity under different circumstances, (ii) the dynamics of decision making, (iii) sequence generation and (iv) the dynamical control of behavior through feedback with the environment. A common solution to such problems is similar to those that have been encountered in physics. In the 60s, laser physics, acoustics and hydrodynamics were investigating similar nonlinear phenomenon - wave synchronization, generation of shock and solitary waves and dynamical chaos. Initially, workers in all three areas tried to solve their problems independently, but after a few years it was clear to everyone that the nonlinear problems they faced were common. As a result, by the beginning of the 70s, we were witness to explosive successes in all of these areas because of seminal developments in the field of nonlinear dynamics of wave systems.

Because of the enormous challenges in understanding the brain as an interacting system of neurons, the same success in such a short time will probably not be possible. However it is clear that a multidisciplinary approach is probably the best way that the linkage between bottom up and top down approaches to understanding the brain's dynamical properties can be established. The underlying assumption is that top down theorizing without experimental confirmation has very limited usefulness because it is not grounded in reality. Similarly, the growing field of biomimetic research that relies solely on engineering principles without taking account of what we know about the anatomy, physiology and dynamics of the nervous system will not be successful.

About fifty years ago M. Rosenblatt, W. Ashby, N. Weiner and other talented medical doctors, physiologists and mathematicians attacked the problem of brain dynamics and possible applications of brain theory from a similar multidisciplinary approach. But because there was insufficient progress in both physiological experiments and nonlinear dynamical theory, their efforts did not provide the necessary foundation for future work. Now, experimental neuroscience and theoretical modeling are ready for a new paradigm. Nowhere is this more clearly illustrated than in efforts to build biomimetic devices for the control of robots and prosthetics as well as to incorporate such devices into living neural circuits. Recent advances at the interface between brain science and engineering have suggested a focal point for this reinvestigation. Electronic neurons have been developed which can form hybrid circuits with biological neurons. Multielectrode recordings from the motor cortex of monkeys have been used to control a robotic arm and the neurotechnology for autonomous robots and prosthetic devices is well under way. We have chosen these topics because a theoretical as well as technological basis is necessary for future work in the area of the so called brain-machine interface and especially important where medical applications are involved. Furthermore the very act of implementing neural systems in simulation or in hardware is actually a useful form of modeling that can lead to a deeper understanding of neuronal information processing. More importantly however, the theoretical foundations for neural systems will only have relevance if they can be modeled both in software and hardware.