Calculator Batch Allocation for Uneven Deep Learning Nodes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In parallel computation using multiple nodes, performance variations due to factors like chip performance and temperature lead to standby times, degrading processing efficiency in deep learning tasks.
Innovation Solution
Distribute the number of batches among nodes based on their measured performance to minimize standby times during result sharing, optimizing batch allocation to maintain continuous processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If parallel computation is performed using multiple nodes, then computation power is improved, but performance variations cause standby times that degrade processing efficiency
Solution Approach 1:
The patent dynamically changes the batch size parameter for each node based on its measured performance characteristics. Fast nodes process larger batches while slow nodes process smaller batches, allowing all nodes to complete their work simultaneously and eliminate standby times, thus resolving the contradiction between utilizing multiple nodes for computation power and maintaining processing efficiency.
2Ease of operation
If uniform batch allocation is used across all nodes, then simplicity is maintained, but performance variations cause standby times during result sharing
Solution Approach 1:
The patent applies local quality by assigning different batch sizes to different nodes based on their individual performance characteristics. Instead of uniform allocation, each node receives a customized batch size that matches its processing capability, eliminating standby times during result sharing while maintaining manageable complexity through automated measurement and allocation.
3Loss of time
If batch sizes are adjusted according to node performance, then standby time is reduced, but measurement and allocation complexity increases
Solution Approach 1:
The patent implements self-service by having each node automatically measure its own performance characteristics and receive appropriate batch size allocations without external intervention. The system autonomously determines optimal batch sizes based on measured performance, reducing standby time while managing complexity through decentralized self-measurement and automated allocation algorithms.
Data Source
AI summary
A deep learning method is performed by a computer having a plurality of nodes. The method comprises measuring, in learning performed by using the plurality of nodes in deep learning, performance of each of the nodes, allocating a number of batches to be processed by an entirety of the plurality of nodes to the individual nodes in accordance with the respective performance measured, and processing the allocated batches in each of the nodes.