Neural Networks with Short-Term Synaptic Dynamics (Leiden, May 23 2008) Misha Tsodyks, Weizmann Institute Mathematical Models of Short-Term Synaptic plasticity Computations in Neural Networks with Short-Term Synaptic Plasticity Experiments: Henry Markram, Eli Nelken Theory: Klaus Pawelzik, Alex Loebel
Multineuron Recording
Principles of Time-Dependency of Release Synaptic Facilitation Synaptic Depression Three Laws of Release Limited Synaptic Resources Release Dependent Depression Release Independent Facilitation AP onset Facilitation Four Parameters of Release Absolute strength Probability of release Depression time constant Facilitation time constant Depression
A Phenomenological Approach to Dynamic Synaptic Transmission 4 Key Synaptic Parameters Absolute strength Probability of release Depression time constant Facilitation time constant
Phenomenological model of synaptic depression (deterministic)
Stochastic transmission .
Binomial model of synaptic release Loebel et al, submitted
The fitting process (deterministic model)
Testing the Model experiment experiment model model
Frequency-dependence of post-synaptic response
Frequency-dependence of post-synaptic response
Frequency-dependence of post-synaptic response
Frequency-dependence of post-synaptic response Tsodyks et al, PNAS 1997
The 1/f Law of Release
Redistribution of Synaptic Efficacy
Frequency Dependence of Synaptic Modification
RSE
Shifting the Distribution of Release Probabilities
Population signal Tsodyks et al, Neural Computation 1998
Population signal
Population signal ‘High-pass filtering’ of the input rate
Steady-state signal
Tsodyks et al PNAS 1997
Synchronous change in pre-synaptic rates
Synchronous change in pre-synaptic rates
Synchronous change in pre-synaptic rates Abbott et al Science 1997
Modeling synaptic facilitation (deterministic) Markram, Wang & Tsodyks PNAS 1998
Facilitation
Frequency-dependence of post-synaptic response
Frequency-dependence of post-synaptic response
Frequency-dependence of post-synaptic response
Population signal Tsodyks et al 1998
Steady-state signal
Signaling with Facilitating Synapses SYNAPSES AS CONTEXTUAL FILTERS
A Small Circuit
Recurrent networks with synaptic depression Tsodyks et al, J. Neurosci. 2000 Loebel & Tsodyks JCNS 2002 Loebel, Nelken & Tsodyks, Frontiers in Neurosci. 2007
Simulation of Network Activity
Simulation of Network Activity
Origin of Population Spikes
Network response to stimulation
Experimental evidence for population spikes DeWeese & Zador 2006
(no inhibition, uniform connections, rate equations) Simplified model (no inhibition, uniform connections, rate equations) i J J
The rate model equations There are two sets of equations representing the excitatory units firing rate, E , and their depression factor, R :
Tone The tonic stimuli is represented by a constant shift of the {e}’s, that, when large enough, causes the network to spike and reach a new steady state
DeWeese, …, Zador 2003
Noise amplitude (Nelken et al, Nature 1999)
Extended model Loebel, Nelken & Tsodyks, 2007
Forward suppression Rotman et al, 2001
Network response to complex stimuli
Network response to complex stimuli
Neural Network Models of Working Memory Misha Tsodyks Weizmann Institute of Science Barak Blumenfeld, Son Preminger & Dov Sagi Gianluigi Mongillo & Omri Barak
Delayed memory experiments – sample to match Miyashita et al, Nature 1988
Persistent activity (Fuster ’73) Miyashita et al, Nature 1988
Neural network models of associative memory (Hopfield ’82) Memories are represented as attractors (stable states) of network dynamics. Attractor = internal representation (memory) of a stimulus in the form of stable network activity pattern Synaptic modifications => Changes in attractor landscape = long-term changes in memory Convergence to an attractor = recall of item from long-term memory into a working form
Associative memory model Hopfield Model (1982): Neurons: Memory patterns: Synaptic connections: Network dynamics:
Associative memory model Hopfield Model (1982): Neurons: Memory patterns: Synaptic modifications:
Associative memory model Hopfield Model (1982): Neurons: Memory patterns: Synaptic modifications:
Associative memory model Hopfield Model (1982): Neurons: Memory patterns: Synaptic modifications: (for each ) if Amit, Guetfriend and Sompolinsky et al 1985
Network dynamics: Convergence to an attractor Overlap Memory #
Network dynamics: Convergence to an attractor Overlap Memory #
Context-dependent representations: gradual change in the pattern Intro – morph sequence memory is a dynamics system that is constantly updated based on experience to capture changes in the environment--- morph example. the representation depends not only on the context in which things are acquired but also on the internal context – what other objects or stimuli are represented in the same system together. Visual stimuli that we are exposed to can change the representation of objects that are already stored in the system. Also suggested by associative memory models
Long-term memory for faces Friends Non Friends … … Preminger, Sagi & Tsodyks, Vision Research 2007
Experiment – Terminology Basic Friend or Non-Friend task (FNF task) Face images of faces are flashed for 200 ms for each image the subject is asked whether the image is a friend image (learned in advance) or not. 50% of images are friends, 50% non-friends, in random order; each friend is shown the same number of times. No feedback is given ? ? ? F/NF F/NF F/NF 200ms 200ms 200ms 200ms 200ms 200ms 200ms 200ms 200ms
Morph sequence Source (friend) Target (unfamiliar) 1 … 20 … 40 … 60 … 80 … 100 Source (friend) Target (unfamiliar)
Morph effect Subject HL -------- (blue-green spectrum) days 1-18 Number of ‘Friend’ responses Bin number Preminger, Sagi & Tsodyks, Vision Research 2007
Morphed patterns – Hopfield model Continuous set of patterns
Morphed patterns – Hopfield model Continuous set of patterns Blumenfeld et al, Neuron 2006
Network dynamics: Convergence to the common attractor Overlap
Network dynamics: Convergence to the common attractor Overlap
Working memory in biologically-realistic networks (D. Amit ‘90)
Working memory in biologically-realistic networks (D. Amit ‘90)
Working memory in biologically-realistic networks (D. Amit ‘90) Brunel & Wang, J. Comp. Neurosci 2001
Working memory in biologically-realistic networks Brunel & Wang, J. Comp. Neurosci 2001
Multi-stability in recurrent networks
Multi-stability in recurrent networks Firing rate (HZ) Synaptic strength (J)
Persistent activity is time-dependent. Rainer & Miller EJN 2002
A Phenomenological Approach to Dynamic Synaptic Transmission 4 Key Synaptic Parameters Absolute strength Probability of release Depression time constant Facilitation time constant 2 Synaptic Variables Resources available (x) Release probability (u) Markram, Wang & Tsodyks PNAS 1998
Synaptic diversity in the pre-frontal cortex Wang, Markram et al, Nature Neuroscience 2006
New idea To hold short-term memories in synapses too, with pre-synaptic calcium level. Use spiking activity only when the memory is needed for processing, and/or to refresh the synapses (smth like ‘rehearsal’). Mongillo, Barak & Tsodyks 2008
Single homogeneous excitatory population – rate model
Gain function
Bifurcation diagram
Bi-stability in networks with synaptic facilitation
Attractors in networks with synaptic facilitation
Integrate and fire network simulations
Integrate and fire network simulations
Integrate and fire network simulations
Robustness and multi-item memory
Robustness and multi-item memory
Random overlapping populations
Predictions Increased pre-synaptic calcium concentration during the delay period, disassociated from increased firing rate Resistance to temporal cessation of spiking Synchronous spiking activity during the delay period Slow oscillations during the delay period (theta rhythm?).
Correlated spiking activity during delay period Constantinidis et al, J. Neurosci. 2001
Theta oscillations and working memory Lee et al, Neuron 2005
Son Preminger (WIS, Rehovot) Dov Sagi (WIS, Rehovot) Barak Blumenfeld (WIS, Rehovot) Omri Barak (WIS, Rehovot) Gianluigi Mongillo (ENS, Paris)