Spike Encoding for Spiking Neural Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing spike encoding methods for spiking neural networks (SNNs) fail to retain maximum information content from real-world analog signals, leading to inefficient pre-processing and suboptimal performance.

Innovation Solution

A method and system that introduce Gaussian noise into input signals, compute mutual information (MI) between noise-introduced signals and encoded spike trains, optimize MI through noise variation, and identify an optimized spike train based on predefined criteria for maximum information retention, which is then fed to SNNs for training.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If conventional encoding techniques (rate coding or temporal coding) are used to convert analog signals to spike trains, then the encoding process is simple and fast, but the information retention is insufficient

Engineering Contradiction:
Improveinformation retentionVSAvoidencoding process complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The method applies preliminary action by introducing Gaussian noise to the input signal before encoding into spike trains. This preprocessing step enhances the information content in the signal, allowing the subsequent simple rate coding or temporal coding to retain more information. The noise is added in advance to improve the signal characteristics prior to the encoding process.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The method employs parameter changes by systematically varying the Gaussian noise parameters (mean and standard deviation) to optimize information retention. By adjusting these parameters and evaluating mutual information metrics, the method identifies optimal noise levels that maximize information preservation during encoding, transforming the fixed-parameter conventional encoding into an adaptive process.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If modern sensor hardware is used to capture real-world stimuli, then the capture quality is improved, but the cost and efficiency are worsened

Engineering Contradiction:
Improvesignal capture qualityVSAvoidencoding efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The method introduces Gaussian noise as an intermediary element that mediates between the captured signal and the encoding process. This intermediary enhancement allows standard sensors to achieve effective information capture without requiring expensive specialized hardware, as the noise addition compensates for limitations in conventional sensor-encoding pipelines.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Loss of information

If the entire spike train length is considered for mutual information computation, then the information measurement is comprehensive, but the computational complexity increases

Engineering Contradiction:
Improveinformation measurement accuracyVSAvoidcomputation complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The method applies segmentation by dividing the spike train into multiple segments or windows for mutual information computation. Instead of computing MI over the entire spike train length at once, the process breaks it into manageable segments, reducing computational complexity while maintaining comprehensive information measurement through aggregated results from all segments.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4009237B1Method and system for optimized spike encoding for spiking neural networks
Publication Date: 2025.07.02 TATA CONSULTANCY SERVICES LTD
  • EP4009237B1 patent drawingFigure 1
  • EP4009237B1 patent drawingFigure 2
  • EP4009237B1 patent drawingFigure 3A

AI summary

This disclosure generally relates optimized spike encoding for spiking neural networks(SNNs). The SNN processes data in spike train format, whereas the real world measurements/input signals are in analog (continuous or discrete) signal format; therefore, it is necessary to convert the input signal to a spike train format before feeding the input signal to the SNNs. One of the challenges during conversion of the input signal to the spike train format is to ensure retention of maximum information between the input signal to the spike train format. The disclosure reveals an optimized encoding method to convert the input signal to optimized spike train for spiking neural networks. The disclosed optimized encoding approach enables maximizing mutual information between the input signal and optimized spike train by introducing an optimal Gaussian noise that augments the entire input signal data.