HTM Union Processor for Stable Temporal Sequence Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing temporal memory systems face challenges in generating a stable representation of temporal sequences and efficiently processing spatial patterns, particularly in maintaining abstraction and invariance over time, which affects their predictive capabilities.
Innovation Solution
The implementation of a processing node that performs spatial pooling using sparse distributed representations and temporal processing, where elements are activated based on selected vectors, and the output remains active longer than a time step, enabling a more stable and invariant representation of temporal sequences through unionization and temporal pooling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If spatial pooling is performed using sparse distributed representations, then the system can efficiently process spatial patterns and reduce computational complexity, but the temporal stability and invariance of the representation may be compromised
Solution Approach 1:
The processing node is divided into distinct functional components: a spatial pooler that performs spatial pooling using sparse distributed representations, and a temporal processing unit that maintains temporal stability. This segmentation allows each component to optimize for its specific function while working together to resolve the contradiction between computational efficiency and temporal stability.
Solution Approach 2:
The patent introduces an intermediary mechanism where the spatial pooler's output is transformed and integrated with temporal information from previous time steps. This intermediary processing layer ensures that the sparse representation maintains both computational efficiency and temporal invariance by combining spatial pooling results with temporal context.
2Productivity
If elements are activated based on selected vectors in sparse distributed representation, then processing speed and efficiency are improved, but the duration of activation and temporal persistence may be reduced
Solution Approach 1:
The temporal processing unit ensures continuous activation of elements by integrating the current sparse distributed representation with previous activation states. This continuity mechanism maintains element activation beyond single time steps, ensuring that useful temporal information is preserved while maintaining high processing speed through efficient sparse operations.
Solution Approach 2:
The system performs preliminary temporal integration by combining current spatial pooling results with previously activated elements before generating the final output. This preliminary action ensures that elements remain active for the required duration while maintaining processing efficiency through optimized sparse matrix operations.
3Reliability
If the output data remains active for a duration longer than the time step, then temporal stability and predictive capability are improved, but the system complexity and memory requirements increase
Solution Approach 1:
The patent merges the spatial pooling function with temporal processing in a unified processing node architecture. By combining these functions rather than implementing them as separate complex systems, the node maintains extended element activation for improved predictive capability while avoiding the complexity overhead of multiple independent components.
Solution Approach 2:
The processing node is designed as a multi-functional unit that simultaneously performs spatial pooling, temporal integration, and active element maintenance. This universal design allows the single node to handle multiple functions that would otherwise require separate specialized components, reducing overall system complexity while maintaining reliable predictive capability.
Data Source
AI summary
Embodiments relate to a processing node of a hierarchical temporal memory (HTM) system with a union processor that enables a more stable representation of sequences by unionizing or pooling patterns of a temporal sequence. The union processor biases the HTM system so a learned temporal sequence may be more quickly recognized. The union processor includes union elements that are associated with incoming spatial patterns or with cells that represent temporal relationships between the spatial patterns. A union element of the union processor may be activated if a persistence score of the union element satisfies a predetermined criterion. The persistence score of the detector is updated based on the activation states of the spatial patterns or cells associated with the detector. After activation, the union element remains active for a period longer than a time step for performing the spatial pooling.