Spiking Neural Sound Classification Using Rate-Synchrony Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional spiking neural networks face challenges in classifying sounds due to the inapplicability of error backpropagation learning methods, requiring feature extraction via MFCC, and limitations in distinguishing sound features using rate codes, which fail to maximize biological advantages.
Innovation Solution
A method and apparatus that preprocess sound data using rate and synchrony codes, perform unsupervised learning with STDP rules, and classify sounds based on neural code propagation characteristics in a spiking neural network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If error backpropagation learning method is used in conventional spiking neural networks, then learning can be performed, but it generates heat from physical hardware and does not meet derivative impossible properties of neurons
Solution Approach 1:
The patent replaces the conventional error backpropagation learning method (which requires derivative operations and generates heat) with Spike Timing-Dependent Plasticity (STDP) learning rules that are biologically plausible for spiking neurons. This substitution eliminates the need for approximations and reduces computational overhead, thereby reducing heat generation from physical hardware while maintaining learning capability.
Solution Approach 2:
The patent changes the learning rule parameters from gradient-based error backpropagation to event-driven STDP rules that operate on spike timing differences. This parameter change makes the learning process compatible with the discrete spike/idle states of neurons, eliminating the derivative operation problem and reducing computational complexity.
2Measurement precision
If MFCC preprocessing is used for sound input, then feature extraction can be performed, but it limits the possibility of extracting biological features and does not maximize biological advantage
Solution Approach 1:
The patent extracts and removes the MFCC preprocessing step from the pipeline, replacing it with direct neural code encoding methods (rate code and synchrony code) that are natively compatible with spiking neural networks. This extraction eliminates the limitation imposed by MFCC and enables the network to directly process and extract biological features from raw audio inputs.
Solution Approach 2:
The patent implements a universal encoding framework using rate codes and synchrony codes that can represent multiple sound features (pitch, volume, timbre) simultaneously without requiring separate preprocessing algorithms. This multi-functional encoding approach maximizes the biological advantage of spiking neural networks by enabling direct extraction of various biological features through the learning process itself.
3Measurement precision
If only rate code encoding is used for sound input, then pitch information can be captured, but it is difficult to distinguish various features of sound input such as volume and timbre
Solution Approach 1:
The patent merges rate code encoding and synchrony code encoding into a unified neural code representation framework. The rate code captures pitch information through firing rates, while the synchrony code captures temporal correlation information for volume and timbre. By combining these two encoding methods, the system preserves all sound features without information loss.
Solution Approach 2:
The patent adds a temporal synchronization dimension to the traditional rate code approach. While rate code operates in the frequency domain (firing rates), synchrony code operates in the temporal domain (spike timing correlations). This dimensional addition enables the network to distinguish multiple sound features simultaneously by encoding them in different temporal patterns.
Data Source
AI summary
A method of classifying sounds based on a neural code in a spiking neural network includes: receiving sounds to be classified and digitally converting the received sounds into sound data; preprocessing the sound data using a multiple neural code-based encoding method including rate code encoding and synchrony code encoding; inputting the preprocessed sound data to a biological spiking neural network to extract features; performing biological spike timing-dependent plasticity (STDP) rule-based learning using the extracted features; and performing classification of the sounds according to neural code propagation characteristics using a test dataset according to a result of the performing of the learning.