Hierarchical Temporal Memory Feedback Grouping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hierarchical Temporal Memory (HTM) systems face challenges in optimal learning of temporal sequences due to incorrect grouping and asynchronous output from nodes, leading to non-optimal or inaccurate learning, and limited computation and storage resources restrict the number of co-occurrences that can be processed.
Innovation Solution
The implementation of a Hierarchical Temporal Memory network with inter-node feedback signals from a parent node to child nodes for grouping patterns and sequences, allowing the parent node to collect forward signals, generate feedback, and guide the child nodes in grouping, merging, splitting, or discarding co-occurrences based on temporal relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If temporal pooling is used to group and store sequences for processing, then the number of sequences can be limited to reduce computational expense, but the learning of temporal sequences may become non-optimal or inaccurate due to incorrect grouping
Solution Approach 1:
The patent implements feedback signals from parent nodes to child nodes that provide information about grouping correctness. Child nodes use this feedback to adjust their grouping of temporal sequences, allowing them to correct grouping errors while maintaining computational efficiency through pooled processing.
Solution Approach 2:
The grouping structure is made dynamic through iterative refinement. Child nodes initially group sequences, receive feedback about grouping accuracy, and then adjust their grouping dynamically. This allows the system to adapt grouping decisions based on learned information rather than using static grouping rules.
2Ease of operation
If child nodes identify and group sequences based on input data, then processing can be distributed, but the child nodes cannot detect problems in their grouping due to lack of external information
Solution Approach 1:
Parent nodes send feedback signals to child nodes that contain information about grouping correctness. This feedback mechanism allows distributed child nodes to detect and correct grouping problems they cannot identify on their own, maintaining both distributed processing and grouping reliability.
Solution Approach 2:
The patent introduces parent nodes as intermediary elements that mediate between the distributed child nodes. These parent nodes aggregate information from multiple child nodes and provide corrective feedback, enabling child nodes to improve their grouping accuracy without centralizing all processing.
3Adaptability or versatility
If nodes at the same level output signals asynchronously due to hardware/software differences, then system flexibility is maintained, but the parent node may identify incorrect sequences
Solution Approach 1:
The patent implements preliminary timing adjustments and synchronization mechanisms before signals reach the parent node. Child nodes prepare their signals with appropriate timing information, allowing the parent node to correctly identify sequences despite asynchronous generation times caused by hardware or software differences.
Data Source
AI summary
A Hierarchical Temporal Memory (HTM) network has at least first nodes and a second node at a higher level than the first nodes. The second node provides an inter-node feedback signal to the first nodes for grouping patterns and sequences (or co-occurrences) in input data received at the first nodes at the first nodes. The second node collects forward signals from the first nodes; and thus, the second node has information about the grouping of the patterns and sequences (or co-occurrences) at the first nodes. The second node provides inter-node feedback signals to the first nodes based on which the first nodes may perform the grouping of the patterns and sequences (or co-occurrences) at the first nodes.