[Comp-neuro] NIPS-Workshop on Echo State Networks and Liquid State Machines

Wolfgang Maass maass at igi.tugraz.at
Sat Oct 14 09:50:53 CEST 2006

You are invited to participate in the

Workshop on Echo State Networks and Liquid State Machines

at NIPS 2006  http://www.nips.cc/Conferences/2006/
(the workshop will take place on Dec. 8 or 9).

There will also be an opportunity to present a poster
at this workshop;  send an abstract or  full paper by Nov. 15 to
Daniela Potzinger <daniela at igi.tugraz.at>.

Organizers :

Dr. Herbert Jaeger, International University Bremen, Germany

Dr. Wolfgang Maass, Technische Universitaet Graz, Austria

Dr. Jose C. Principe, University of Florida, USA

Motivation, Goals and Details of this Workshop:

A new approach to analyzing and training recurrent neural networks
(RNNs) has emerged over the last few years. The central idea is to
regard a sparsely connected recurrent circuit as a nonlinear,
excitable medium, which is driven by input signals (possibly in 
conjunction with feedbacks from readouts). This recurrent circuit is 
--like a kernel in Support Vector Machine applications-- not adapted 
during learning. Rather, very simple (typically
linear) readouts are trained to extract desired output signals. Despite 
its simplicity, it was recently shown that such simple networks have (in 
combination with feedback from readouts) universal computational power, 
both for digital and for analog computation. There are currently two 
main flavours of such networks. Echo state networks were
developed from a mathematical and engineering background and are 
composed of simple sigmoid units, updated in discrete time. Liquid state 
machines were conceived from a mathematical and computational 
neuroscience perspective and usually are made of
biologically more plausible, spiking neurons with a continuous-time 

These approaches have quickly gained popularity because of their 
simplicity, expressiveness, ease of training. In addition they provide a 
new perspective for modeling cortical computation that differs in 
several aspects from previous models. Generic cortical microcircuits are 
seen from this perspective as explicit implementations of
kernels (in the sense of SVMs), raising the question how such explicit 
kernels can be optimized by unsupervised learning procedures for a 
particular inputs statistics and a particular range of computational tasks.

Quite a number of researchers have started to work on this approach, and
a first special issue of a journal (Neural Networks) dedicated to this 
topic is currently assembled. Furthermore results of neurobiological 
experiments that test predictions of this approach have just been 
completed, and further experiments are currently in the
planning stage.

The goals of this workshop are to

--provide a resume of the current state of knowledge, in particular
regarding theory and results of firsts experimental tests of its 
predictions in neuroscience

--discuss consequences of this approach for computational and 
theoretical neuroscience

--to guide future research by working out the essential open problems

--to encourage new applications (e.g. in reinforcement learning, speech 
processing, handwriting recognition reading, auditory processing).

The target audience consists of neuroscientists, cognitive scientists, 
theoreticians, neural network researchers, and engineers.


The workshop will begin with 3 mini-tutorials (20 minutes each) on

-- Theory of ESNs and LSMs

-- Resulting perspectives for neuroscience research

-- How to design a reservoir or liquid for particular tasks.

The rest of the morning session, and the first part of the afternoon 
session will be devoted to presentations of the most exciting new 
research results in this area (format: 10-12 talks of lengths between 15 
and 25 minutes, followed each by 5-10 minutes of discussion).
The last 60 minutes of the workshop will be devoted to a discussion of
open problems, and resulting new strategies for experimental planning 
and data-analysis in neuroscience.

Prof. Dr. Wolfgang Maass
Institut fuer Grundlagen der Informationsverarbeitung
Technische Universitaet Graz
Inffeldgasse 16b ,   A-8010 Graz,  Austria
Tel.:  ++43/316/873-5811
Fax   ++43/316/873-5805

More information about the Comp-neuro mailing list