Spiking Neural Network Synapse Weight Attenuation for Information Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current spiking neural networks lack a forgetting characteristic, leading to inefficient information processing and storage, as they cannot effectively discard irrelevant information, resulting in large storage needs and low processing efficiency.
Innovation Solution
Implementing a signal processing method that uses multiple orders of synapse weight parameters with different attenuation rates over time, simulating a forgetting mechanism by adjusting weights based on learning rules and stimulation effects, allowing unimportant information to be forgotten and important features to be retained.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a spiking neural network stores all received information without forgetting, then information storage completeness is improved, but information storage space increases and processing efficiency decreases
Solution Approach 1:
The patent applies parameter changes by introducing multiple orders of synapse weight parameters (first-order, second-order, etc.) that decay at different rates. Each order represents a different timescale of memory retention, allowing the system to automatically differentiate between important and unimportant information based on decay patterns rather than selective storage
Solution Approach 2:
The patent segments the synapse weight into multiple orders (first-order synapse weight parameter, second-order synapse weight parameter, etc.), where each order handles different timescales of information retention. This segmentation allows parallel processing of information at different importance levels without requiring centralized selection mechanisms
2Loss of information
If a spiking neural network stores all received information without forgetting, then information storage completeness is improved, but processing efficiency decreases
Solution Approach 1:
By changing the decay rate parameters of different synapse weight orders, the system dynamically adjusts how long different types of information are retained. Fast-decaying higher-order parameters filter out transient noise while slow-decaying lower-order parameters preserve important patterns, improving processing efficiency without losing critical information
Solution Approach 2:
The patent introduces dynamic decay mechanisms where synapse weight parameters automatically adjust their retention characteristics over time. The decay rates are determined by the order of the synapse weight, creating a dynamic filtering system that adapts to different information patterns without requiring external control
3Productivity
If multiple orders of synapse weight parameters with different decay rates are used, then forgetting characteristic is improved and storage requirements are reduced, but device complexity increases
Solution Approach 1:
The synapse weight is segmented into multiple orders, where each order is handled by existing neural network components. This segmentation approach maintains biological plausibility while implementing complex forgetting characteristics, as each order can be processed by standard spiking neuron mechanisms
Solution Approach 2:
The patent makes the synapse weight parameter multi-functional by having it serve both as a connection strength indicator and as a timescale selector. The same synapse weight structure handles both information retention and decay control, reducing the need for separate mechanisms and minimizing additional complexity
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
A signal processing method and apparatus based on a spiking neural network are disclosed. The method includes: determining a first signal F1(t) output by a first neuron (S110); processing the first signal F1(t) by using q orders of synapse weight parameters wq(t),wq-1(t),···,wi(t), to obtain a second signal F2(t), where a speed at which an initial function Wx+10t met by an (x+1)th-order synapse weight parameter of the q orders of synapse weight parameters attenuates with time t is higher than a speed at which an initial function Wx0t met by an xth-order synapse weight parameter attenuates with the time t, q is a positive integer greater than 1, and 1 ≤ x ≤ q-1 (S120); and inputting the second signal F2(t) to a second neuron, where the second neuron is a next-layer neuron of the first neuron (S130). In this method, an input signal is processed by using a plurality of orders of synapse weight parameters, and initial functions of the orders of the synapse weight parameters attenuate with time at different speeds, so that a forgetting characteristic of the neural network is simulated, the neural network is more consistent with reality, it is easy to implement feature extraction, and a problem of an excessively large information storage amount can be resolved.