Neural Temporal Coding With Relative Delay Lines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for neural temporal coding struggle to effectively distinguish and learn temporal patterns beyond mere coincidence or order of inputs, requiring biologically-inspired yet reduced complexity solutions for coding, decoding, and recognizing temporal spike signal patterns.
Innovation Solution
The method employs a relative delay line abstraction to delay synaptic inputs, applies a dynamic spiking model to determine neuron behavior based on weighted and delayed inputs, and adjusts weights using an unsupervised learning rule to capture timing relations, enabling the recognition of long and large spatial-temporal patterns through hierarchical multi-layer neural networks with self-connected synapses and dendritic delays.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex temporal pattern recognition methods are used, then recognition precision is improved, but device complexity increases
Solution Approach 1:
The patent divides complex temporal patterns into smaller sub-patterns that can be processed by individual neurons with specific delay lines. Each neuron handles a segment of the temporal pattern by detecting specific spike timing relationships, and the overall pattern is recognized through integration of these segmented detections across the neural network.
Solution Approach 2:
The patent introduces temporal delay dimensions to the neural network architecture by adding delay lines to synapses. This transforms the problem from recognizing temporal patterns in spike timing to recognizing spatial patterns in delayed signal arrivals, allowing standard spatial pattern recognition mechanisms to handle temporal coding.
2Adaptability or versatility
If biologically-inspired neural models are used, then learning capability is improved, but computational complexity increases
Solution Approach 1:
The patent implements self-organizing maps and Hebbian learning rules that allow the neural network to automatically adapt its weight configurations and delay line settings based on input patterns. The system performs unsupervised learning by detecting temporal correlations in spike trains and automatically configuring synapses to recognize recurring temporal patterns without external supervision.
Solution Approach 2:
The patent uses learning rules that dynamically adjust synaptic weights and delay parameters based on temporal spike patterns. The learning process modifies connection strengths and timing parameters to optimize pattern recognition performance, allowing the network to adapt to different temporal coding schemes and pattern types.
3Measurement precision
If hierarchical multi-layer networks are used, then pattern matching capability is improved, but processing time increases
Solution Approach 1:
The patent pre-configures delay lines and synaptic weights during the learning phase to encode temporal pattern templates. When recognition is needed, the network immediately compares incoming spike patterns against these pre-configured templates using the predetermined delay relationships, enabling rapid pattern matching without sequential processing.
Solution Approach 2:
The patent implements feedback mechanisms where the output of one neural layer provides delayed feedback to earlier layers, allowing the network to refine pattern recognition decisions through iterative comparison. This feedback loop enables the network to detect temporal patterns across multiple time scales and improve recognition accuracy through repeated evaluation.
Data Source
AI summary
Certain aspects of the present disclosure support a technique for neural temporal coding, learning and recognition. A method of neural coding of large or long spatial-temporal patterns is also proposed. Further, generalized neural coding and learning with temporal and rate coding is disclosed in the present disclosure.