Probabilistic approaches to the understanding of inference and learning in the cerebral cortex

NII Shonan Meeting:

@ Shonan Village Center, Dec. 4-8, 2011 CANCELLED

Organizers:

  • Haruo Hosoya (Dept of Computer Science, University of Tokyo)
  • Florian Roehrbein (Albert Einstein College of Medicine, Yeshiva University)

 

Overview

The cerebral cortex is an intriguing biological structure that realizes sophisticated intelligence by enormous yet regularly organized circuits. Recent trends in theoretical investigations regard the cortex as graph structure performing some kind of probabilistic computation. In particular, on top of graphical models such as Bayesian networks, Markov random fields, and deep belief networks, various inference and learning algorithms have been proposed and compared with neuroanatomical and neurophysiological data. The four-day meeting aims to at bringing together researchers from both neuroscience and machine learning communities, to foster the exchange of results and opinions and to discuss future directions.

The cerebral cortex is an intriguing biological structure that
realizes sophisticated intelligence by enormous yet regularly
organized circuits. Understanding its function and neural
implementation strategy is one of the most important goals in
computational neuroscience.

One recent direction suggested by several theoretical investigators is
to interpret the cerebral cortex as a hierarchical graph structure
performing some kind of probabilistic computation. In particular, a
number of inference and learning models have been proposed on the
basis of Bayesian networks, Markov random fields, and deep belief
networks, which are then compared with neuroanatomical and
neurophysiological data.

However, while these investigations have given some valuable insights,
numerous questions are still under active research. These include:

  • which graphical models are suitable abstractions of the cortical
    computation?
  • which inference and learning algorithms are performed by the cortex?
  • what are the theoretical properties of such algorithms?
  • what physiological properties can such algorithms reproduce and predict?
  • which neural circuits can implement such algorithms?
  • what relationships are there between neural circuits and neuroanatomy?
  • what are possible applications of such models?

The four-day meeting aims at bringing together neuroscience and
machine learning researchers from all over the world in order to
foster the exchange of ideas and opinions and the discussion for
future directions.