Computational Neuroscience Colloquium

Computational Neuroscience Connection



UW Computational Neuroscience Colloquium

Room G417, Health Sciences Building, UW

Past talks

Thursday, April 28, 3.00-4.00

Why don’t we understand V2?
J. Anthony Movshon

HSB G318


Tuesday, Nov 23th 11:30AM - 1:00 PM (double header of short talks)

* Information Maximization Fails to Maximize Expected Utility in a Simple Foraging Model*


Information theory has successfully explained the organization of many biological phenomena,
from the physiology of sensory receptive fields to the variability of certain DNA sequence
ensembles. Some scholars have proposed that information should provide the central explanatory
principle in biology, in the sense that any behavioral strategy that is optimal for an organism’s
survival must necessarily involve efficient information processing. Here we challenge this view
by providing a counterexample. We present an analytically tractable model for a particular in-
stance of a perception-action loop: a creature searching for a wandering food source confined to
a one-dimensional ring world. The model incorporates the statistical structure of the creature’s
world, the effects of the creature’s actions on that structure, and the creature’s strategic decision
process. The underlying model takes the form of a Markov process on an infinite dimensional
state space. To analyze it we construct an exact coarse graining that reduces the model to a
Markov process on a finite number of “information states”. This mathematical technique allows
us to make quantitative comparisons between the performance of an information-theoretically
optimal strategy with other candidate search strategies on a food gathering task. We find that

1. Information optimal search does not necessarily optimize utility (expected food gain).

2. The rank ordering of search strategies by information performance does not predict their
ordering by expected food obtained.

3. The relative advantage of different strategies depends on the statistical structure of the
environment, in particular the variability of motion of the source.

We conclude that there is no simple relationship between information and utility. Behavioral
optimality does not imply information efficiency, nor is there a simple tradeoff between the two
objectives of gaining information about a food source versus obtaining the food itself.


* Soft Clustering Decoding of Neural Codes *


Methods based on Rate Distortion theory have been successfully used to
cluster stimuli and neural responses in order to study neural codes at a
level of detail supported by the amount of available data. They
approximate the joint stimulus-response distribution by soft-clustering
paired stimulus-response observations into smaller reproductions of the
stimulus and response spaces. An optimal soft clustering is found by
maximizing an information-theoretic cost function subject to both
equality and inequality constraints, in hundreds to thousands of

The method of annealing has been used to solve the corresponding high
dimensional non-linear optimization problems. The annealing solutions
undergo a series of bifurcations in order to reach the optimum. We study
that system using bifurcation theory in the presence of symmetries. The
optimal models found by distortion methods have symmetries: any soft
clustering data can lead to another equivalent model simply by permuting
the labels of the classes. These symmetries are described by S_N, the
algebraic group of all permutations on N symbols. The symmetry of the
bifurcating solutions is dictated by the subgroup structure of S_N. In
this presentation we describe these symmetry breaking bifurcations in
detail, and indicate some of the consequences stemming from the form of
the bifurcations.


TUES Oct 13, 1:00-2:00. Research talk by Christopher Honey, Princeton.

Hierarchies and History-dependence in the Human Brain

To organize and understand events unfolding in our everyday environment, it is necessary that we integrate information over seconds and minutes of time. This information integration occurs effortlessly and at multiple time-scales in parallel. We propose that this aspect of human cognitive function relies on a hierarchically organized network of brain regions in which distinct levels of the hierarchy integrate information over distinct time scales. Primary sensory regions are hypothesized to perform little temporal integration while heteromodal association and frontal areas integrate information over longer time periods. I will first review some recent fMRI work aimed at measuring history dependence in the visual and auditory systems, and will then present some results from an ECOG project conducted at NYU Medical Center, in which we examine how distinct frequency components of human neurophysiological recordings may differentially represent temporal contextual information.

FRI APRIL 30, 3:30-4:30. Research talk by Rajesh Rao, UW.

Learning to act under uncertainty: A neural model based on partially observable Markov decision processes.

Animals are faced with the problem of choosing actions and making
decisions on the basis of uncertain sensory information and incomplete
knowledge of the environment. We apply the framework of partially
observable Markov decision processes (POMDPs) to this problem and
suggest a neural implementation in the networks of the cortex and the
basal ganglia. We illustrate the model using the random dots motion
discrimination task and show that the model can learn to solve this
task by maintaining beliefs over states and maximizing future expected
reward. After learning, belief computation in the model resembles LIP
responses while the reward prediction error signal shares some
similarities with dopamine responses in the basal ganglia.

THURS MAY 20. Nancy Kopell, Boston U. gives UW Applied Math Boeing Seminar.
4:00 PM, Guggenheim Hall, room TBA.

FRI FEBRUARY 19, 3:30-4:30. Discussion of recent back-to-back Science papers on correlated spiking (Renart et al, Ecker et al., Science Vol. 327. no. 5965, 2010). Eric Shea-Brown + member of Fairhall Lab presenting.

FRI MARCH 12, 3:30-4:30. Research talk by Greg Schwartz, UW.

Nonlinear Computation in the Retina

Retinal ganglion cell responses are often modeled as a linear receptive field followed by the nonlinearity of spike generation. I will discuss three types of computations in the retina that defy explanation in such a framework. While the mechanisms of some of these phenomena are better understood than others, I hope to convey an appreciation of some of the sophisticated visual processing that can take place within the circuitry of the retina. A number of themes in this work will provide links to general principles of computation in sensory systems.


(organizers: Adrienne Fairhall, Eric Shea-Brown)