Reservoir-Based Spiking Neural Network for Edge Time-Series Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional time-series classification techniques are computationally intensive, making it challenging to achieve low-latency real-time responses on computationally constrained edge devices, which is crucial for applications like predictive maintenance and IoT systems that require efficient processing of sensor data.

Innovation Solution

A reservoir-based spiking neural network is implemented for time-series classification, using spike encoders to encode time-series data into spike trains, which are then processed by a spiking reservoir to extract spatio-temporal features and train a classifier for real-time classification, leveraging event-based asynchronous processing and sparse data handling.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional time-series classification techniques (CNN, RNN, SVM, k-NN) are used, then classification accuracy is improved, but computational complexity increases making real-time processing on edge devices infeasible

Engineering Contradiction:
Improveclassification accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments the classification task into two parts: an offline training phase that creates a classification model, and an online inference phase that runs efficiently on edge devices. The reservoir computing model separates feature extraction (online, low-complexity) from feature learning (offline, high-complexity), enabling real-time processing with maintained accuracy

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary action by pre-training the reservoir computing model offline using comprehensive datasets. The classification model is prepared in advance with learned features and parameters, so that during real-time operation on edge devices, only lightweight inference is required, reducing online computational complexity while maintaining accuracy

Inventive Principle:
Principle #10Preliminary action

2Power

If data is transferred to cloud for processing, then computational power is sufficient, but latency and power consumption increase

Engineering Contradiction:
Improveprocessing capabilityVSAvoidlatency
Core Design Contradiction:
PowerVSLoss of time

Solution Approach 1:

The system extracts the essential classification functionality into a compact model that can be deployed on edge devices. By taking out only the necessary inference capabilities from the full training system, the solution enables local processing that eliminates data transfer latency while consuming minimal power on edge devices

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The reservoir computing model acts as an intermediary between raw sensor data and classification results. It processes data locally on the edge device, serving as a mediator that eliminates the need for continuous cloud communication, thereby reducing latency and power consumption associated with data transfer

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4270250A1Methods and systems for time-series classification using reservoir-based spiking neural network
Publication Date: 2023.11.01 TATA CONSULTANCY SERVICES LTD
  • EP4270250A1 patent drawingFigure 1
  • EP4270250A1 patent drawingFigure 2A
  • EP4270250A1 patent drawingFigure 2B

AI summary

The present disclosure relates to methods and systems for time-series classification using a reservoir-based spiking neural network, that can be used at edge computing applications. Conventional reservoir based SNN techniques addressed either by using non-bio-plausible backpropagation-based mechanisms, or by optimizing the network weight parameters. The present disclosure solves the technical problems of TSC, using a reservoir-based spiking neural network. According to the present disclosure, the time-series data is encoded first using a spiking encoder. Then the spiking reservoir is used to extract the spatio-temporal features for the time-series data. Lastly, the extracted spatio-temporal features of the time-series data is used to train a classifier to obtain the time-series classification model that is used to classify the time-series data in real-time, received from edge devices present at the edge computing network.