Graph Convolutional Network Training via Supernode Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to efficiently train and deploy Graph Convolutional Networks (GCNs) on large-scale graph data due to computational complexity and scalability challenges.
Innovation Solution
The method involves compressing graphs into supernodes and superedges, using a GCN model to generate modified embedding information based on supernode information, embedding information, and connection information, and iteratively updating the compressed graph to manage memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional computing architectures are used to train GCNs on large-scale graphs, then the model can be trained, but the training time is prolonged and scalability is limited
Solution Approach 1:
The patent applies segmentation by dividing the large-scale graph into multiple partitions or blocks that can be processed independently and in parallel. This allows the GCN training to be distributed across multiple computing units, significantly reducing training time and improving scalability while maintaining the integrity of graph convolution operations across partition boundaries.
2Quantity of substance
If the graph size is increased to handle larger datasets, then the model coverage is improved, but the computational complexity increases
Solution Approach 1:
The patent segments the large graph into manageable partitions, allowing the system to handle larger graphs by distributing the computational load. Each partition can be processed with reduced complexity, and the overall system achieves linear or near-linear scaling rather than exponential complexity growth.
Solution Approach 2:
The patent introduces a new dimension of parallel processing by utilizing multiple computing units or GPUs simultaneously. This dimensional expansion allows the system to handle larger graphs by adding computational resources in parallel rather than sequentially, effectively managing computational complexity through spatial distribution of work.
3Power
If more computational resources are allocated to handle large-scale graphs, then the processing capability is improved, but the system cost and resource requirements increase
Solution Approach 1:
The patent segments the computational workload into discrete partitions that can be distributed across available resources. This allows efficient utilization of existing computational resources without requiring a proportional increase in hardware, as each partition can be processed independently with moderate resource requirements.
Solution Approach 2:
The patent designs a universal processing framework that can handle different graph sizes and structures using the same partitioning and processing mechanisms. This multi-functional approach allows the system to efficiently utilize computational resources across various workloads without requiring specialized hardware for each scenario, reducing overall resource requirements.
Data Source
AI summary
Embodiments of the present disclosure describe a processing apparatus that may obtain embedding information of a first node to be added to a graph and connection information between the graph and the first node, receive, supernode information for a supernode of a compressed graph corresponding to the graph, wherein the supernode includes a plurality of nodes from the graph, and generate modified embedding information of the first node based on the supernode information, the initial embedding information, and the connection information using a graph convolutional network (GCN) model.