Local and Global Gating of

Synaptic PlasticityManuel A. Sanchez-Montañes

Hannes Schulz

University of Osnabrück, Department of Cognitive Science

Action and Cognition II / May 2nd 2005

Learning on Global vs. Local Scale

Mechanisms for learning in neuronal nets:

Local

Hebb’s Rule & modifications

Forms representations on

cortex

Global

Signals to the whole network

Modifies local learning

Learning Goals

Local Learning:

Representation of all

stimuli

Possibility to add new

stimuli later on

Independence from

presentation frequency

Global Learning:

Influence size of

representation

(→ experimental

evidence on basal

forebrain)

◮ How to connect both?

The Neuronal Model Experimental Results Discussion

Outline

1 The Neuronal Model

The Cell

The Network

The Inputs

2 Experimental Results

3 Discussion

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Outline

1 The Neuronal Model

The Cell

The Network

The Inputs

2 Experimental Results

3 Discussion

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

The Cell Model

∑

integrate-and-fire

refractory period

delayed transmission

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Network Topology

Input Cells Excitatory Cells

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Network Topology

Input Cells Excitatory Cells Inhibitory Cells

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Network Topology

Input Cells Excitatory Cells Inhibitory Cells

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Network Topology

Input Cells Excitatory Cells Inhibitory Cells

Nucleus Basalis

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Network Topology

Input Cells Excitatory Cells Inhibitory Cells

Nucleus Basalis

Layers 1&2, 3&2 fully connected

Random weight initialization

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Network Topology

Input Cells Excitatory Cells Inhibitory Cells

Nucleus Basalis

Layers 1&2, 3&2 fully connected

Random weight initialization

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Retrograde Signal Inhibition

Retrograde Signal

Dendrit

Firing

Cell

Inhibitory Signal

◮ Retrograde signal

of firing cell

blocked by

inhibitory inputs

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Heterosynaptic Long Term Depression

Presynaptic Cell Postsynaptic Cell

“Normal” Condition

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Heterosynaptic Long Term Depression

Presynaptic Cell Postsynaptic Cell

“Normal” Condition

LTD

Heterosynaptic LTD:

postsynaptic cell

active

presynaptic cell

inactive in time

window

◮ Synaptic efficacy

decrease

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Relative Timing in Hebb Learning

Sample interpretations of “Synchronous activity”:

Symmetric coincidence

window.

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Relative Timing in Hebb Learning

Sample interpretations of “Synchronous activity”:

Symmetric coincidence

window.

Asymmetric

coincidence window.

Used in this study.

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Input

Trial I

Ten different stimuli

pseudorandom order

500 examples shown to

network

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Input

Trial I

Ten different stimuli

pseudorandom order

500 examples shown to

network

Trial II

9+1 different stimuli

pseudorandom order

500 examples shown to

network

One stimulus paired with

stimulus in basal-neuron

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion The Cell The Network The Inputs

Input

Trial I

Ten different stimuli

pseudorandom order

500 examples shown to

network

Trial II

9+1 different stimuli

pseudorandom order

500 examples shown to

network

One stimulus paired with

stimulus in basal-neuron

Both run 40 times with random start parameters

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion

Outline

1 The Neuronal Model

The Cell

The Network

The Inputs

2 Experimental Results

3 Discussion

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion

Formed Representations – Trial I

Neuron Specificity

44.8% unspecific

50.5% specific to 1 stimulus

4.7% intermediate

Stimuli shown more often

to network not better

represented

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion

Formed Representations – Trial II

Number of neurons

representing

paired stimulus

increases

Number of neurons

representing other

stimuli stay the

same

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion

Role of the Global Mechanism

1 Basal ganglion neuron fires

2 Inhibitory neurons have a

prolonged refractory period

3 Inhibitory neurons do not fire

4 Excitatory neurons can fire

and shorten refractory period

again

Network Topology

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion

Role of the Global Mechanism

◮ Unchanged mean activity of

inhibitory neurons

◮ Delay of inhibitory activity relative to

excitatory activity

◮ Retrograde APs invade dendritic

tree

Network Topology

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion

The Same, in Graphs

Delay of inhibitory

activity relative to

excitatory activity

Number of neurons

representing paired

stimulus increases

Number of neurons

representing other stimuli

stay the same

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion

Outline

1 The Neuronal Model

The Cell

The Network

The Inputs

2 Experimental Results

3 Discussion

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion

Comparison of Results to Goals

New stimuli can be trained w/o loss

Optimally activated neurons inhibit othersUnspecific neurons are result

Frequency invariance, variable representation size

Invariance due to mechanism described aboveStimulus representation enhanced by globalmechanism

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The Neuronal Model Experimental Results Discussion

Predictions for Experimentors

During basal forbrain stimulation:

Delayed inhibitory activity

Invasion of dendritic tree by more retrograde APs

Hannes Schulz Local and Global Gating of Synaptic Plasticity

The End