Hierarchical Temporal Memory Action Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hierarchical Temporal Memory (HTM) systems fail to effectively incorporate actions into the learning process, limiting their ability to understand and perceive the world by only considering temporal aspects of sensed input data without accounting for the actions that modify these inputs.
Innovation Solution
The implementation of HTM networks that utilize spatio-temporal sensed input data associated with actions to infer causes and learn from sequences of sensory inputs, allowing the system to recognize patterns and actions governing these inputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If HTM systems only consider temporal aspects of sensed input data, then the system complexity remains manageable, but the perception and understanding capability is incomplete
Solution Approach 1:
The system segments the learning process into two distinct modules: a sequence learner that processes temporal sequences of sensed inputs, and an action learner that processes actions and their effects. This segmentation allows each module to specialize in one aspect (temporal or action-based) while working together to achieve complete perception, resolving the contradiction by making the system both versatile and manageable in complexity.
Solution Approach 2:
The patent adds a new dimension to HTM processing by incorporating action-based learning alongside temporal sequence learning. Instead of only processing temporal sequences, the system now processes actions (causes) and their effects on sensed inputs, creating a multi-dimensional learning framework that enhances perception capability while maintaining structured organization.
2Measurement precision
If HTM systems incorporate actions into the learning process, then the learning accuracy and causality understanding improve, but the computational requirements and processing time increase
Solution Approach 1:
The system performs preliminary processing by separating action identification from sequence processing. The action learner first identifies actions and their effects on sensed inputs, creating a structured representation that the sequence learner then processes. This preliminary action reduces the computational burden during sequence processing, improving learning accuracy while managing processing time through staged computation.
3Loss of information
If the system processes both spatial and temporal patterns with action associations, then the causal inference capability improves, but the data processing complexity increases
Solution Approach 1:
The patent introduces an intermediary structure (action representation layer) that mediates between raw sensed inputs and the learning process. Actions serve as intermediaries that link causes to effects on sensed inputs, allowing the system to process causal information systematically. This intermediary structure organizes complex data relationships, improving causal inference while managing processing complexity through structured representation.
Data Source
AI summary
A set of sequences of sensed input patterns associated with a set of actions is generated by performing at least a first action on data derived from a real-world system. A subset of the sequences of sensed input patterns that form a group associated with the first action is determined. A new sequence of sensed input patterns is received. A first value which indicates the probability that the new sequence of sensed input patterns is associated with the first action based on the subset of sequences of sensed input patterns is determined and stored in a memory associated with the computer system.