Stochastic Delay Plasticity for Spiking Neural Network Synapses
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing spike-timing dependent delay plasticity implementations in neural networks result in ever-increasing delays when postsynaptic spikes occur after a group of presynaptic spikes, leading to gratuitous delay and saturation, as there is no mechanism to minimize overall delay after presynaptic spikes have been clustered.
Innovation Solution
The introduction of stochastic delay plasticity, where the sign of delay change is probabilistically altered based on the time difference between presynaptic and postsynaptic spikes, allowing for three possible outcomes (positive, negative, or no change) with specific probabilities, enabling more efficient updates and improved network performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deterministic delay plasticity is used to align presynaptic and postsynaptic spikes, then spike alignment is improved, but synaptic delays become excessively large and saturate when postsynaptic spikes occur after presynaptic spike clusters
Solution Approach 1:
The patent transforms the static deterministic delay adjustment into a dynamic stochastic process. Instead of fixed delay changes based on spike timing differences, the system employs probabilistic delay modifications where the magnitude and direction of delay changes vary stochastically. This allows the network to explore different delay configurations and escape from local minima that cause excessive delay accumulation, while still achieving effective spike alignment through the statistical tendency of the stochastic updates.
2Stability of the object's composition
If positive delay changes are applied to cluster presynaptic spikes, then spike clustering is improved, but overall system delay increases and saturates
Solution Approach 1:
The patent changes the parameter of delay adjustment from deterministic to stochastic. By introducing probability distributions for delay modifications, the system can achieve spike clustering through the aggregate effect of many small stochastic adjustments rather than large deterministic changes. This prevents the runaway positive feedback that causes delay saturation, as the stochastic nature introduces variability that prevents consistent accumulation in one direction.
Data Source
AI summary
A method of operating a spiking neural network having neurons coupled together with a synapse includes monitoring a timing of a presynaptic spike and monitoring a timing of a postsynaptic spike. The method also includes determining a time difference between the postsynaptic spike and the presynaptic spike. The method further includes calculating a stochastic update of a delay for the synapse based on the time difference between the postsynaptic spike and the presynaptic spike.