Nearest Neighbor Graph Construction via Distributional Approximation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing graph machine learning technologies face challenges in efficiently constructing accurate nearest neighbor structures for tabular data, particularly in reducing computational burden and improving prediction accuracy without the need for explicit graph construction.
Innovation Solution
A method that determines cross-class neighborhood similarities using a theoretical approximation, allowing for the selection of a suitable nearest neighbor graph based on inter-class and intra-class similarities, and builds a nearest neighbor graph without requiring explicit graph construction, leveraging a Hierarchical Naïve Bayes model and Monte Carlo simulations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If explicit graph construction is performed using traditional nearest neighbor methods, then prediction accuracy can be improved, but computational burden and resource consumption increase significantly
Solution Approach 1:
The patent changes the parameter representation from explicit graph structures to distributional parameters (mean vectors and covariance matrices) that characterize the data. By applying linear transformations to these parameters rather than manipulating explicit graph structures, the method achieves comparable prediction accuracy with significantly reduced computational burden, directly resolving the contradiction between measurement precision and productivity
Solution Approach 2:
The patent creates a simplified copy of the graph structure in the form of distributional parameters (Gaussian distributions characterized by mean and covariance). Instead of working with the complete explicit graph construction, the method uses these parameter copies to capture essential structural information, enabling efficient computation while maintaining prediction accuracy
2Reliability
If explicit graph construction is performed to capture neighborhood relationships, then classification performance improves, but memory consumption and computational resources increase
Solution Approach 1:
The patent transforms the representation from explicit graph edges and nodes (which consume significant memory) to compact distributional parameters (mean vectors and covariance matrices). This parameter transformation maintains the ability to capture neighborhood relationships for reliable classification while dramatically reducing the quantity of data structures that must be stored in memory
Solution Approach 2:
The patent extracts only the essential characteristics of the graph structure (neighborhood relationships) in the form of distributional parameters, separating these essential features from the complete explicit graph construction. This extraction allows the method to retain classification performance while eliminating the memory burden of storing and processing full graph structures
Data Source
AI summary
A method for construction of nearest neighbor structures includes determining a set of cross-class neighborhood similarities based on a set of distributions of data obtained by applying a model to data present in a dataset. The method selects a first cross-class neighborhood similarity from the set of cross-class neighborhood similarities based on one or more inter-class cross-class neighborhood similarities and one or more intra-class cross-class neighborhood similarities, and builds a nearest neighbor graph based on the first cross-class neighborhood similarity. The present invention can be used in a variety of applications including, but not limited to, several anticipated use cases in drug development, material synthesis, and medical/healthcare.