Selective Data Retention in Grounding Based Applications
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current memory management techniques for logical neural networks (LNNs) are inefficient due to high memory requirements, particularly during upward and downward inference passes, where data generated by the upward pass often exceeds what is needed for the subsequent downward pass, leading to unnecessary data storage and limited memory utilization.
Innovation Solution
Implementing a memory optimization technique that discards data generated by the upward inference pass and reconstructs it only when necessary, using snapshot layers to store delta data and discarding temporary data from segment layers, allowing for efficient memory usage and reduced memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data generated by the upward inference pass is retained for the subsequent downward pass, then inference accuracy is maintained, but memory requirements increase significantly
Solution Approach 1:
The patent applies the discarding and recovering principle by selectively discarding data generated during the upward inference pass that is not required for the downward pass, while recovering only the essential data needed for accurate inference. This reduces memory usage by eliminating unnecessary data retention while maintaining inference accuracy through selective data recovery.
Solution Approach 2:
The patent extracts and retains only the critical data elements from the upward pass that are necessary for the downward pass, separating essential information from redundant data. This extraction approach maintains inference reliability by preserving necessary data while reducing memory requirements by excluding unnecessary data.
2Reliability
If all data from upward pass is stored for downward pass, then complete inference information is available, but memory utilization becomes inefficient
Solution Approach 1:
The system discards redundant data from the upward pass and recovers only the minimal necessary data for complete inference, thereby improving memory utilization efficiency while maintaining inference completeness through selective data management.
Solution Approach 2:
The patent creates selective copies of only the essential data portions needed for the downward pass, rather than copying all upward pass data. This copying strategy ensures inference completeness with necessary data while improving memory utilization by avoiding redundant data duplication.
Data Source
AI summary
An embodiment includes configuring a grounding based application (GBA) structure comprising a plurality of layers, where the plurality of layers comprises a first layer and a second layer, the first layer having a first node that receives an input associated with a child node responsive to an input query, and the second layer having a second node that outputs a response to the input query. The embodiment also includes evaluating, using the GBA structure, a logical inference based on the input query, where the evaluating comprises generating a first truth table associated with the first node, where the generating of the first truth table comprises retaining truth values resulting from a downward inference pass on the GBA structure and discarding truth values resulting from an upward inference pass on the GBA structure. The embodiment also includes outputting, responsive to the evaluating, an output truth value representative of the logical inference.