Workshop on "Cognitive Dynamics in Neural Systems: Mathematical and Computational Modeling",
Lyon, France, March 29-30

Program

Rationale

Suggestive topics

Invited speakers

Background

Format

Program board

 

Rationale

Animals are proof that complete cognitive systems can be realized in neural substrates. It is thus natural that engineers from AI and machine learning have tried to design advanced cognitive systems on the basis of artificial neural networks. This has led to illuminating concepts and architectures in fields like computational linguistics, dynamic pattern recognition, autonomous agents, or evolutionary robotics. However, if one takes a close and critical look, one finds that nowhere do artificial systems close to biological levels of performance.  One important cause for this gap is a lack of appropriate mathematical concepts. Biological neural systems are high-dimensional, nonlinear, heterogeneous, multiscale, nonstationary, stochastic, and heavily input-driven - a cocktail of properties which overwhelms current dynamical systems theory. Inasmuch as we do not possess mathematical models for such systems, we cannot understand them; and inasmuch as we do not understand, we cannot engineer.

This workshop will bring together researchers from three fields:

  • cognitive scientists, roboticists and machine learning engineers who develop complex, neural-network-based architectures;
  • computational neuroscientists who apply existing methods from dynamical systems theory to neural dynamics;
  • mathematicians who work on extensions of dynamical systems theory in directions that appear relevant for neurodynamics and complex neural learning architectures.

The purpose of the workshop is to create an awareness, across the three fields, of the general and fundamental methodological challenges that oppose easy progress; and at the same time, an appreciation of partial solutions, questions, and methods in the respective other fields which may inspire progress in one's own.

 

Suggestive topics

  • design principles for neural robot/agent  architectures
  • dynamical correlates of rules, concepts and symbols
  • (neuro)dynamical models of memory and attention
  • integrating top-down and bottom-up pathways in complex neural architectures
  • multiscale dynamics in modular or hierarchical neural systems
  • levels of description and analysis: from spikes to rates to fields
  • agent-environment coupling dynamics
  • new concepts from the young field of non-autonomous dynamical systems
  • new concepts from the young field of dynamics on networks

(go to top)

Invited speakers

(go to top)

Background

The workshop is funded through the ORGANIC FP7 European project (http://organic.elis.ugent.be), and is indeed the culmination event of this project which ends shortly after the workshop. The ORGANIC project targetted recurrent neural network (RNN) architectures for speech and handwriting recognition, couched in the reservoir
computing paradigm. The project was successful, reaching or sometimes surpassing conventional state-of-the-art performance with an entirely RNN-based approach, demonstrating the viability of RNN based large-scale learning architectures. In the course of ORGANIC research a number of remarkable dynamical phenomena were trained into RNNs, which, as it turned out, defy current concepts of dynamical systems theory. These observations will be shared with the participants in one of the workshop sessions. By staking out the theoretical grounds for "nonstandard" phenomena in cognitive dynamics, the workshop appropriately leads from the terminating ORGANIC project into research challenges that arise beyond, and through, the investigations completed in this project. (go to top)

Format

This will be a workshop, with ample time for discussion both in the plenum and during breaks. The latter will be more generously timed than is usually the case, in order to foster personal, informal conversations. The scientific input to the workshop will come from 9 invited presentations (see "invited talks" button in the menu), plus an introduction from the organizers. Besides these no further oral presentations are planned. In order to allow young researchers to display their work and establish personal contacts, a 2-hr poster session will be held early during the workshop, accessible during the break times later on. Poster abstracts (see Submission page for details) are to be submitted before the workshop for an informal review by the organizers. There will be no proceedings, but the organizers will write a "lessons learnt"  document after the event which will be placed online. It will be decided during the workshop whether this summary document should be expanded into a publication with joint authorship. (go to top)

Program Board

The program board is composed of the principal investigators of the ORGANIC project:

(go to top)