Distributed Synchrony: a model for cortical communication

1
Distributed Synchrony a model for cortical
communication

• Madhur Ambastha
• Jonathan Shaw
• Zuohua Zhang
• Dana H. Ballard
• Department of Computer Science
• University of Rochester
• Rochester, NY

Dana Ballard - University of Rochester 1
2
Summary
1. There is a computational hierarchy.2. At
the bottom of the hierarchy is the need to
calibrate.3 . To communicate
throughout cortex quickly, calibration uses
the g band
3
(No Transcript)
4
(No Transcript)
5

1. Computational Timescales
Context Select a set of active behaviors 10s
Resource Map active behaviors onto motor system .3s
Routines update state information 100ms
Calibration represent sensory/motor/reward 20ms
Computational quanta 2ms
6
2. How can the Cortical Memory Self-Calibrate?
Olshausen and Field 97 Rao and Ballard 99
7
Code Input I with synapses U and output r
Min E(U,r) I-Ur2 F(r) G(U)
Coding cost of residual error
Coding Cost of model
8
Synapses are Trained with Natural Images
1. Apply Image
2. Change firing
3. Change Synapses
9
An Example LGN-V1Circuit
e I - Ur
rest
r

-
10
Hierarchical Memory Organization
Fellerman and Van Essen 85
11
A Slice Through The Cortex
X
LGN
V1
V2
12
Endstopping
RF
Rao and Ballard, Nature Neuroscience 1999
13
3. Can Predictive Coding work with individual
spikes?
14
Spike Timing Model

_
r
Loop delay - 20 milliseconds
15
LGN-V1 Circuit using Spikes
rest
e

-
r
rest
e

-
16
Spike Models
Spike is probabilistic
Deterministic spike has area
17
LGN ON
LGN OFF
I
Ur
I-Ur
prediction error
input
feedback prediction
18
Coding Cells
Orientation Distribution
Receptive Fields
19
Responses are Random and Phasic
20
Projection Pursuit
I
u1
r1 I u1
r2 ( I - r1 u1 ) u2
r1
u2
r2
21
Microcircuit Details
r1 I u1
r2 I u2 - r1 u1 u2
1
-r1 u1 u2
2
22
(No Transcript)
23
Summary 1Distributed Synchrony is motivated by
fourprinciple constraints
1. Fast, reliable intercortical
communication2. The need for a cell to
multiplex3. Need to poll the input4 .The need
to reproduce observed cell responses
24
Summary 2Isolating Computations The Binding
problem
Solutions1. There is no binding problem -
Movshon2. Fast weight changes at synapses - von
der Malsburg3 .Synchrony encodes the stimulus -
Singer4 .Synchrony encodes the answer - Koch
and others 5 .Synchrony encodes the process -
Distributed Synchrony
25
Thanks !
26
Handling the Error Term with Predictive Coding
LGN
Cortex
27
Roelfsema et al PNAS 2003
28
Synchronous Spikes Can Propagate
Diesmann, Gewaltig,Aertsen Nature 402, p529 1999
29
Minimum Description Length - Bayesian Version
MaxM P(MD) MaxMP(DM)P(M)/P(D)
Can neglect P(D) and take logs
MaxMlog P(DM) log P(M)
Or equivalently minimize negative logs
MinM - log P(DM) - log P(M)
Coding cost of residual error
Coding cost of model
If we use exponentiated probability
distributions, log cancels negated exponent so
30
Singer group, J Neuroscience 1997
31
Cortical Inhibitory Cells Can Oscillate at 20-50
Hz
Beierlein, Gibson, Connors Nature Neuroscience
3 p904 2000
32
Temporal Rate CodingA Strategy that cannot
possibly work
33
(No Transcript)
34
LGN ON
low
LGN OFF
Reconstruction as a function of Coding Cost
LGN ON
high
LGN OFF
35
Spectral software supplied by Daeyeol Lee
36
Distributed Synchrony
37
Coding Cost as a function of Signaling Strategy
38
Axonal Propagation Speeds Evidence?
2-6 cm/s
0.1 - 0.4 cm/s
39
Visual Routine
40
(No Transcript)
41
Reverse Correlation

D


42
Spatio-temporal behavior of LGN Cells
Using Reverse Correlation
Experiment (Reid Usrey)
Model