Spike Neural Network Anomaly Detection Circuit
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in quickly detecting anomaly data input while continuously processing data, especially in environments where normal data ranges change, requiring a low-power solution that can adapt and quickly identify anomalies.
Innovation Solution
An anomaly data detection device utilizing a spike neural network (SNN) that generates spike signals from input data, determines output spike firings based on synaptic weights, and adjusts these weights through evolutionary learning to detect anomalies, incorporating a sampler and detection circuit to generate signals based on spike rates and neuron firings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional anomaly detection methods are used to continuously process input data, then detection accuracy may be maintained, but power consumption increases and response speed decreases
Solution Approach 1:
The patent transforms continuous analog input data into discrete spike signals through parameter changes (analog-to-digital conversion, spike encoding). This discretization enables event-driven processing where computations only occur when anomalies are detected, dramatically reducing power consumption while maintaining detection speed. The spike neural network processes only relevant events rather than continuously processing all input data.
Solution Approach 2:
The sampler component implements periodic action by sampling input data at specific time intervals and generating session data periodically. This periodic sampling approach allows the system to process data in discrete time windows rather than continuously, reducing overall power consumption while maintaining adequate detection capability for time-varying anomalies.
2Reliability
If adaptive learning is implemented to track changes in normal data ranges, then detection accuracy improves, but computational complexity and processing time increase
Solution Approach 1:
The spike neural network performs preliminary learning actions during idle periods or between anomaly detections, updating synaptic weights based on incoming data to adapt to changing normal ranges. By preparing the model in advance and using incremental learning rather than full retraining, the system maintains high detection reliability without significant processing time penalties during actual anomaly detection events.
Solution Approach 2:
The system implements self-service through autonomous adaptive learning where the neural network automatically adjusts its synaptic weights and detection thresholds based on incoming data patterns. This self-adaptation eliminates the need for manual recalibration and allows the system to maintain high reliability in dynamic environments without external intervention or extensive processing overhead.
3Reliability
If continuous data processing is performed to ensure no anomalies are missed, then detection completeness improves, but power consumption and processing load increase
Solution Approach 1:
The system extracts only the essential features from continuous input data by converting them into spike signals that capture the most salient information. The sampler and spike encoder extract critical data points while discarding redundant information, allowing the neural network to process a compressed representation that maintains detection completeness with significantly reduced energy consumption compared to processing the full continuous data stream.
Data Source
AI summary
Disclosed is an anomaly data detection device, which includes a sampler that generates session data including first to m-th sample data based on input data input during a first time interval, a spike signal generator that generates first to m-th spike signals respectively corresponding to the first to m-th sample data based on the session data, a spike neural network that detects whether an output spike fires in at least one output neuron from among output neurons based on the first to m-th spike signals and synaptic weights of each of the output neurons, and a detection circuit that generates a detection signal based on the number of output neurons firing the output spike, and each of the first to m-th spike signals is generated by converting feature information of the corresponding first to m-th sample data into a spike rate code.