GNN Pre-processing Acceleration via COO to CSR Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Graph neural network (GNN) pre-processing consumes most of the time in GNN operations, hindering efficiency and automation.
Innovation Solution
The method involves converting an original graph from a coordinate list (COO) format to a compressed sparse row (CSR) format, generating a sub-graph by reducing the graph's degree through uniform random sampling, and creating an embedding table corresponding to the sub-graph, utilizing a conversion unit, a sub-graph generation unit, and an embedding table generation unit.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If GNN pre-processing is performed using traditional methods, then the graph can be processed, but the processing time is excessively long and efficiency is low
Solution Approach 1:
The patent divides the pre-processing workflow into distinct modular units: a conversion unit that transforms COO format graphs to CSR format, a sub-graph generation unit that creates reduced-degree sub-graphs, and an embedding table generation unit that produces embedding tables. This segmentation allows each unit to be optimized independently and processed in parallel, significantly improving pre-processing efficiency while reducing overall time consumption.
2Loss of information
If the original graph is processed in full detail, then complete information is retained, but computational overhead and processing complexity increase significantly
Solution Approach 1:
The patent extracts only the essential components needed for GNN processing by generating sub-graphs with reduced degree from the original graph. The sub-graph generation unit selectively retains critical graph structures and relationships while removing redundant elements, thereby maintaining sufficient information for accurate GNN operations while dramatically reducing computational complexity and processing requirements.
3Extent of automation
If manual pre-processing methods are used, then flexibility is maintained, but automation and speed are compromised
Solution Approach 1:
The patent implements self-service automation where the conversion unit automatically transforms input graphs from COO to CSR format without manual intervention. The sub-graph generation unit autonomously creates reduced-degree sub-graphs, and the embedding table generation unit automatically produces the required embedding tables. This fully automated pipeline eliminates manual pre-processing steps while maintaining high processing speed and consistency.
Data Source
AI summary
Provided is an apparatus for accelerating graph neural network (GNN) pre-processing, the apparatus including a conversion unit configured to convert an original graph in a coordinate list (COO) format into a graph in a compressed sparse row (CSR) format, a sub-graph generation unit configured to generate a sub-graph with a reduced degree of the graph in the CSR format, and an embedding table generation unit configured to generate an embedding table corresponding to the sub-graph.