Hierarchical Neural Network for Temporal Pattern Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current artificial neural networks (ANNs) using unsupervised learning struggle with efficiently recognizing temporal patterns due to inefficient cataloging of coincidences and noise corruption, leading to disorganized data and difficulty in identifying rare temporal sequences.
Innovation Solution
A hierarchical system with modules that process input streams into population codes, using filter functions and Hebbian learning rules to cluster patterns, enabling efficient temporal sequence learning and error correction, and employing spike-timing dependent plasticity for recurrent connections to predict future events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If unsupervised learning is used to avoid human labeling labor, then ease of operation is improved, but pattern recognition performance deteriorates
Solution Approach 1:
The system segments the input stream into discrete time bins, dividing the continuous temporal data into manageable segments. This allows the network to process and recognize temporal patterns systematically without requiring supervised labeling, resolving the contradiction between ease of operation and pattern recognition performance.
Solution Approach 2:
The system performs preliminary actions by pre-processing the input stream to create a population code representation before temporal pattern recognition. This preliminary encoding organizes the data structure in advance, enabling efficient unsupervised learning and improving pattern recognition performance without human intervention.
2Difficulty of detecting and measuring
If coincidences are cataloged to identify temporal patterns, then pattern recognition capability is improved, but device complexity increases due to high-dimensional input space
Solution Approach 1:
The system transforms the high-dimensional input stream into a lower-dimensional population code representation by mapping temporal patterns onto a discrete set of time bins. This dimensional transformation reduces the complexity of cataloging coincidences while preserving temporal pattern information, enabling efficient pattern recognition without overwhelming computational complexity.
3Measurement precision
If temporal sequences are searched in high-dimensional input streams, then pattern detection accuracy is improved, but productivity decreases due to extremely low probability of coincidences
Solution Approach 1:
The system performs preliminary action by pre-encoding the input stream into a population code that highlights temporal structures before searching for patterns. This preliminary organization concentrates the search into relevant temporal windows, dramatically improving productivity while maintaining pattern detection accuracy by filtering out low-probability coincidences.
Solution Approach 2:
The system maintains continuity of useful action through recurrent connections that persist temporal pattern representations across time steps. This continuous reinforcement of temporal structures enables efficient pattern detection without repeatedly scanning the entire high-dimensional input space, thereby improving productivity while maintaining accuracy.
Data Source
AI summary
The present invention is directed to a system and methods by which the determination of pattern recognition may be facilitated. More specifically, the present invention is a system and methods by which a plurality of computations may be conducted simultaneously to expedite the efficient determination of pattern recognition.