Graph Learning with Wild Card Nodes for Indefinite Elements
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep Tensor lacks a learning pattern for graph structures without node labels, leading to determination errors when dealing with indefinite node elements, as it cannot correctly discriminate data with partial structures containing unknown nodes.
Innovation Solution
The learning program generates extended graph data by adding wild card nodes to represent indefinite elements, performing tensor decomposition and deep learning using a neural network to improve discrimination accuracy for graphs with indefinite elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If learning with labels is performed in Deep Tensor, then node label values are utilized for discrimination, but determination errors occur when specific node labels are indefinite or not present in training data
Solution Approach 1:
The patent introduces a label embedding layer as an intermediary component that maps node labels to continuous vector representations. This embedding layer enables the model to handle indefinite or unseen labels by representing them in a continuous space, allowing generalization beyond discrete training labels while maintaining discrimination accuracy through the learned embedding vectors.
Solution Approach 2:
The patent transforms discrete node label parameters into continuous embedding vector parameters. By changing the parameter representation from discrete labels to continuous vectors, the model gains the ability to handle indefinite labels through vector interpolation and similarity measurements, resolving the contradiction between precision and adaptability.
2Adaptability or versatility
If learning without labels is performed in Deep Tensor, then indefinite node elements are allowed, but determination errors occur when partial structures with other labels are incorrectly matched
Solution Approach 1:
The label embedding layer serves as a mediator that prevents incorrect matching by computing meaningful distances between label vectors. Even when node labels are indefinite, the embedding space provides a structured representation that distinguishes between different partial structures, preventing false matches while maintaining adaptability.
Solution Approach 2:
The patent introduces dynamic label embeddings that can adapt their representations during learning. The embedding vectors are learned dynamically to capture the semantic relationships between labels, enabling the model to dynamically adjust its discrimination behavior based on the specific graph structure and label combinations encountered.
3Ease of manufacture
If Deep Tensor uses fixed learning patterns, then implementation is straightforward, but it cannot handle both labeled and unlabeled graph structures with indefinite elements
Solution Approach 1:
The patent creates a universal learning framework that combines labeled and unlabeled learning capabilities in a single model architecture. The label embedding layer and modified loss function work together to enable the model to handle both labeled and unlabeled graph structures with indefinite elements, providing multi-functionality while maintaining implementation simplicity through a unified approach.
Data Source
AI summary
A non-transitory computer-readable recording medium stores a learning program that causes a computer to execute a machine learning process for graph data. The machine learning process includes: generating, from graph data to be subjected to learning, extended graph data where at least some of nodes included in the graph data have a value of the nodes and a value corresponding to presence or absence of an indefinite element at the nodes; and obtaining input tensor data by performing tensor decomposition of the generated extended graph data, performing deep learning with a neural network by inputting the input tensor data into the neural network upon deep learning, and learning a method of the tensor decomposition.