Information%20encoding%20and%20processing%20via%20spatio-temporal%20spike%20patterns%20in%20cortical%20networks%20Misha%20Tsodyks,%20Dept%20of%20Neurobiology,%20Weizmann%20Institute,%20Rehovot,%20Israel

1
Information encoding and processing via
spatio-temporal spike patterns in cortical
networksMisha Tsodyks, Dept of Neurobiology,
Weizmann Institute, Rehovot, Israel

• Thanks to
• Alex Loebel, Omri Barak, Asher Uziel, Henry
  Markram

2
Rate coding (V1)
3
Y. Prut, , M. Abeles 1998
4
W. Bair C. Koch 1996
5
DeWeese, , Zador 2003
6
Open questions
How do precise spike patterns emerge in the
cortex? How can they be robust in the presence
of random firing of surrounding neurons? What is
the relation between the spike patterns and the
stimuli that they are coding for? How can the
information carried by spike patterns
be processed?
7
Open questions
How do precise spike patterns emerge in the
cortex? How can they be robust in the presence
of random firing of surrounding neurons? (Synfire
chains? I dont like it!) What is the relation
between the spike patterns and the stimuli that
they are coding for? How can the information
carried by spike patterns be processed?
8
Recurrent networks with dynamic synapses
(unstructured)
Tsodyks et al 2000
9
Wang Yun et al 1998
10
Modeling Time-Dependent Release

• 4 Synaptic Parameters
• Absolute strength
• Probability of release
• Depression time constant
• Facilitation time constant

11
Population spikes
12
Population spikes
13
Origin of Population Bursts
14
Temporal Correlations
15
Network response to stimulation
16
Simplified model (no inhibition, uniform
connections, rate equations)
i
J
J
17
The rate equations

• Two sets of equations representing the excitatory
  units firing rate, E, and their depression
  factor, R

Loebel Tsodyks 2002
18
Population spikes in the simplified model
19
Adiabatic approximation
(except during the population spike)
20
Adiabatic approximation
(except during the population spike)
Population spike
21
Adiabatic approximation
Population spike
Higher spontaneous activity lower propensity
for population spikes.
22
Response to excitatory pulses

• Inputs

Response
Population spike
No population spike
23

• Inputs

Response
Population spike
No population spike
24

• Inputs

Response
Population spike
No population spike
25
Response to tonic stimuli

• The tonic stimuli is represented by a
  constant shift of the es, that, when large
  enough, causes the network to burst and reach a
  new steady state

26
(No Transcript)
27
(No Transcript)
28
Interaction between stimuli
29
Open questions
How do precise spike patterns emerge in the
cortex? (Synfire chains?) How can they be robust
in the presence of random spontaneous and evoked
firing of surrounding neurons? What is the
relation between the spike patterns and the
stimuli that they are coding for? How can the
information carried by spike patterns
be processed?
30
Extended model
Loebel Tsodyks 2006
31
The model response to a pure tone
32
Constraining the propagation of the PS along the
map
33
Forward suppression
Rotman et al, 2001
34
Network response to complex stimuli
35
Network response to complex stimuli
36
Open questions
How do precise spike patterns emerge in the
cortex? (Synfire chains?) How can they be robust
in the presence of random spontaneous and evoked
firing of surrounding neurons? What is the
relation between the spike patterns and the
stimuli that they are coding for? How can the
information carried by spike patterns
be processed?
37
Processing spike patterns Tempotron (Guetig and
Sompolinsky, 2006)
Learned patterns
vs background patterns
Barak Tsodyks, 2006
38
Variance-based learning
where
39
Cost function for learning
40
Learning rules for spatio-temporal patterns
Gradient descent
Correlation-based
41
Convergence of learning
42
Performance of the tempotron
43
Measuring the tempotron performance
44
Robustness to time warps
45
Conclusions

• 1. Networks with synaptic depression can
  encode spatio-temporal inputs by precise spike
  patterns.
• 2. Random spontaneous activity could play a
  crucial role in setting the sensitivity of the
  network to sensory inputs (top-down control,
  attention, expectations, ?)
• 3. Coding by spike patterns is highly nonlinear.
• 4. Effective learning rules for recognition of
  spike patterns in tempotron-like networks can be
  derived.