Hierarchical Mode-Wise Pooling for Accurate Feature Vector Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for hierarchical message passing in graph neural networks and table representation learning using flat mode structures result in machine learning inefficiencies and inaccurate outcomes.
Innovation Solution
A hierarchical mode-wise pooling technique that transforms hierarchically-indexed sets of feature vectors, utilizing the entire hierarchy of modes at every level of granularity to improve machine learning accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional hierarchical message passing or flat mode structures are used, then the system is simpler to implement, but machine learning accuracy deteriorates
Solution Approach 1:
The patent segments the feature vector transformation process into distinct hierarchical levels (pooling, transformation, unpooling operations at different granularity levels). This segmentation allows the system to process features at multiple scales simultaneously, improving accuracy while maintaining manageable complexity through modular organization of operations
Solution Approach 2:
The patent implements nested hierarchical structures where feature vectors are processed at multiple levels of granularity, with each level containing and contributing to the next. The pooling operations aggregate features at coarser levels while unpooling operations distribute them back to finer levels, creating a nested processing architecture that enhances accuracy without linearly increasing complexity
2Measurement precision
If more training samples are used to improve accuracy, then machine learning accuracy improves, but training time increases
Solution Approach 1:
The patent creates transformed copies of feature vectors through pooling and unpooling operations, generating multiple representations of the same underlying data at different hierarchical levels. This copying mechanism allows the model to learn from diverse feature representations without requiring additional training samples, thereby maintaining accuracy while reducing training time
Solution Approach 2:
The patent transforms feature vectors by changing their hierarchical parameters through pooling operations that aggregate features at different granularities. This parameter transformation enables the model to extract more information from the same data by varying the level of abstraction, improving accuracy without increasing sample size or training duration
Data Source
AI summary
An approach is provided for processing received feature vectors whose indices are in an original hierarchical index space. A transformation of each mode of the feature vectors is generated and includes a pooling operation, a transformation operation, and an unpooling operation. The pooling operation aggregates and collapses the feature vectors to produce a coarse-grained feature set having a collapsed index space. The transformation operation transforms the coarse-grained feature set into a transformed feature set that preserves the collapsed index space. The unpooling operation generates a final feature set in the original hierarchical index space by reverting pooling from the transformed feature set. The feature vectors are transformed without any pooling or unpooling to produce an additional set of transformed feature vectors. A summation that preserves the original hierarchical index space is generated by summing the final feature set and the additional set of transformed feature vectors.