Spiking Neural Network Membrane Potential Regularization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The behavior of circuit models in spiking neural networks does not consistently match the actual circuit behavior, leading to reduced analysis accuracy due to discrepancies in membrane potential and spike signal timing.

Innovation Solution

A multilayer spiking neural network with an additive computing portion that suppresses the lower limit of membrane potential in each layer through learning, aligning the circuit model operation with actual circuit operation by optimizing weight coefficients and membrane potential regularization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If virtual circuit models based on Kirchhoff's rule are used to realize neuron models, then the neural network can be implemented, but discrepancies arise between the behavior of the virtual circuit model and that of the actual circuit, reducing analysis accuracy

Engineering Contradiction:

Improveconsistency between circuit model and actual circuitVSAvoidanalysis accuracy

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

The patent applies parameter changes by modifying the membrane potential calculation to account for circuit-specific characteristics. Specifically, it adjusts the lower limit of membrane potential and incorporates circuit operation characteristics into the learning process, allowing the virtual model to adapt its parameters to match actual circuit behavior more closely.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent implements feedback mechanisms through learning processes that adjust weight coefficients based on the difference between virtual circuit model predictions and actual circuit behavior. The system uses error signals from comparison between modeled and actual membrane potentials to refine the model parameters iteratively.

Inventive Principle:

Principle #23Feedback

2Adaptability or versatility

If the lower limit of membrane potential is not suppressed, then the neuron model can operate freely, but negative membrane potential fluctuations occur that cause discrepancies with actual circuit operation

Engineering Contradiction:

Improvemembrane potential rangeVSAvoidconsistency with actual circuit

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent applies preliminary anti-action by proactively suppressing negative membrane potential fluctuations before they can cause significant discrepancies. The learning process pre-adjusts the lower limit of membrane potential and weight coefficients to prevent unrealistic negative values that would diverge from actual circuit behavior.

Inventive Principle:

Principle #9Preliminary anti-action

Solution Approach 2:

The system dynamically adjusts the membrane potential parameter range through learning, modifying the lower limit to prevent excessive negative fluctuations while maintaining sufficient adaptability for neural computation. This parameter optimization ensures the model stays within realistic circuit operating ranges.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20240403617A1Computing device, neural network system, neuron model device, computation method, and trained model generation method

Publication Date: 2024.12.05 NEC CORP

AI summary

A computing device includes: a multilayer spiking neural network including a plurality of neurons. and includes an additive computing portion in which a lower limit of membrane potential of the neurons in each layer is suppressed by learning.