Wave-based STDP Synaptic Weight Update via Spike Phase Differential
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural network learning techniques, such as Synaptic Time Dependent Plasticity (STDP), are inefficient and time-consuming, especially in spiking neural networks, due to the need for extensive processing cycles that involve both learning and unlearning, making them wasteful and slow to converge.
Innovation Solution
The implementation of a modified STDP-based learning method, referred to as Wave-based STDP (WSTDP), which updates synaptic weights based on the spiking rates and phase differences between spike trains, allowing for more efficient weight adjustments and improved learning convergence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional STDP learning techniques are used in spiking neural networks, then synaptic weights can be adjusted through learning cycles, but the process becomes time-consuming and inefficient due to extensive processing cycles required for both learning and unlearning
Solution Approach 1:
The patent changes the parameters used for synaptic weight adjustment from traditional STDP timing-based rules to a hybrid approach combining rate-based modulation with phase differential signaling. This parameter change allows the network to converge faster by using phase information to guide learning direction, reducing the number of processing cycles needed while maintaining learning accuracy
Solution Approach 2:
The invention introduces periodic phase modulation to the spike trains, where synapses are updated based on the phase differential between pre- and post-synaptic spikes within each periodic cycle. This periodic structure organizes the learning process into efficient cycles, reducing the overall training time compared to continuous conventional STDP processing
2Reliability
If extensive processing cycles are used for learning and unlearning in conventional neural networks, then comprehensive weight adjustment is achieved, but computational resources are wasted and convergence is slow
Solution Approach 1:
The patent implements feedback mechanisms where the phase differential information from spike trains is continuously monitored and used to modulate synaptic weights. This feedback loop allows the network to adaptively adjust weights based on temporal relationships, achieving reliable learning with fewer processing cycles by eliminating redundant learning-unlearning iterations
Solution Approach 2:
The invention applies preliminary phase-based modulation to spike trains before full learning cycles begin. By pre-processing spike trains with phase information extraction and using this to initialize or guide subsequent weight adjustments, the network achieves faster convergence while maintaining learning reliability, reducing the need for extensive processing cycles
Data Source
AI summary
Techniques and mechanisms for determining the value of a weight associated with a synapse of a spiking neural network. In an embodiment, a first spike train and a second spike train are output, respectively, by a first node and a second node of the spiking neural network, wherein the synapse is coupled between said nodes. The weight is applied to signaling communicated via the synapse. A value of the weight is updated based on a product of a first value and a second value, wherein the first value is based on a first rate of spiking by the first spike train, and the second value is based on a second rate of spiking by the second spike train. In another embodiment, the weight is updated based on a product of a derivative of the first rate of spiking and a derivative of the second rate of spiking.