Associative Memory Circuit Using Echo State Network for Pattern Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional associative memory devices face challenges in natural language processing due to the need for separate hardware for time-series signal recognition and static pattern recognition, leading to increased hardware resources and combinatorial explosion of computational complexity during learning.
Innovation Solution
An associative memory device utilizing an echo state network with an input layer, intermediate layer, and output layer, which learns and processes memory patterns to output recall patterns, thereby avoiding the combinatorial explosion and enabling high-speed information processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If separate hardware is used for time-series signal recognition and static pattern recognition, then both functions can be performed, but hardware resources increase
Solution Approach 1:
The patent combines time-series signal recognition and static pattern recognition into a single Echo State Network architecture. The reservoir layer processes both dynamic temporal patterns and static patterns uniformly, eliminating the need for separate hardware systems while maintaining both functional capabilities.
Solution Approach 2:
The Echo State Network is designed as a universal computing architecture that can handle multiple types of pattern recognition tasks through its reservoir layer. By using a single system with adjustable parameters, the network achieves multi-functionality for both time-series and static pattern analysis without requiring dedicated hardware for each function.
2Adaptability or versatility
If recurrent neural circuit model is used for associative memory learning, then various pattern recalls can be created, but computational complexity explodes during learning convergence
Solution Approach 1:
The reservoir layer is pre-configured with fixed random weights before learning begins. This preliminary setup creates a rich dynamic system that can perform various pattern recalls, while the actual learning only requires adjusting the output layer weights, dramatically reducing computational complexity during the learning convergence process.
Solution Approach 2:
The patent extracts the complex learning problem from the entire recurrent network by separating the reservoir layer (with fixed weights) from the output layer (with learnable weights). This extraction allows the system to maintain rich pattern recall capabilities while reducing the computational burden to only the output layer training, avoiding combinatorial explosion.
Data Source
AI summary
An associative memory device includes a memory pattern specifying unit that stores a fixed-length numeric vector as a memory pattern; a recall pattern specifying unit that stores a numeric vector associated with the memory pattern as a recall pattern; an associative memory learning unit that acquires an associative memory circuit through learning by using an echo state network including an input layer, an intermediate layer, and an output layer. The values of all units constituting the input layer are input in parallel to the intermediate layer. The values are output in parallel from the intermediate layer through neural network processing performed within the intermediate layer set as values of all units constituting the output layer. The recall pattern is output in parallel from the associative memory circuit.