Neural Network Graph Partitioning for Compiler Search Space Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional artificial neural network compilers using machine learning-based search technology face challenges in detecting optimal intermediate representation due to exponentially increasing search spaces with larger networks, and they fail to simultaneously consider the use efficiencies of various resources.
Innovation Solution
The method involves partitioning the artificial neural network graph into partial graphs using an artificial neural network partitioning module, determining a partial partitioning space for intensive search using a partitioning space minimization module, and generating node information to select an optimal partitioning method, ultimately producing intermediate representations optimized for resource use across different computing hardware resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning-based search technology is used to search for optimal intermediate representation, then resource use efficiency is improved, but search space exponentially increases when artificial neural network size increases
Solution Approach 1:
The patent segments the artificial neural network graph into multiple partial graphs using a graph partitioning module. This segmentation divides the large search space into smaller, manageable subspaces that can be searched more efficiently. By partitioning the network graph into communities or clusters, the compiler can perform targeted searches within each partition rather than exhaustively searching the entire network, thus reducing the exponential search space complexity while maintaining resource use efficiency optimization.
2Productivity
If conventional machine learning-based search technology is used, then optimal intermediate representation is derived for specific resource use efficiency, but use efficiencies of various kinds of resources are not simultaneously and complexly considered
Solution Approach 1:
The patent implements a multi-objective optimization framework that simultaneously considers multiple resource types including computational resources, memory resources, and energy consumption. The intermediate representation search evaluates candidates based on multiple objectives rather than a single resource metric, making the compiler versatile across different hardware platforms and resource constraints. This multi-functional approach allows the same compilation process to optimize for various resource efficiencies concurrently.
Data Source
AI summary
A method of compiling an artificial neural network includes partitioning a graph representing an artificial neural network into partial graphs, determining a partial partitioning space which has to be intensively searched for in a total partitioning space and then generating node information about nodes associated with the partial partitioning space, in a case which defines, as the total partitioning space, a set including the number of all cases corresponding to a partitioning method capable of being used in a process of partitioning the graph into the partial graphs, partitioning the graph into the partial graphs by using an optimal partitioning method selected by configuring and searching for the partial partitioning space, based on the node information, and generating artificial neural network intermediate representations, based on the partial graphs, and arranging the artificial neural network in different computing hardware resources based on the artificial neural network intermediate representations.