Distributed Learning Broadcast Scheduling for Lower Radio Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing iterative learning processes in decentralized machine learning systems are resource-inefficient due to each agent entity broadcasting its model parameter vector to neighboring entities during each global iteration, leading to excessive radio resource consumption.
Innovation Solution
A method and coordinator entity are introduced to selectively choose a subset of agent entities to broadcast their local model parameter vectors based on utility metrics, such as centrality, data importance, and availability, reducing the number of broadcasts and optimizing resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all agent entities broadcast their local model parameter vectors during each global iteration, then the learning convergence is ensured, but the radio resource consumption increases excessively
Solution Approach 1:
The patent segments the set of all agent entities into multiple subsets, where only agents in selected subsets broadcast their model parameter vectors during each global iteration. This segmentation approach maintains learning convergence by ensuring sufficient information exchange while reducing radio resource consumption by excluding agents from broadcasting in subsequent iterations.
Solution Approach 2:
The patent implements periodic action by rotating which subsets of agents broadcast in each global iteration. The coordinator entity selects different subsets for different iterations, creating a periodic broadcasting pattern that ensures all agents eventually broadcast (maintaining convergence) while no single agent broadcasts continuously (reducing energy consumption).
2Productivity
If all agent entities broadcast their local model parameter vectors during each global iteration, then the model updates are fully exchanged, but the energy efficiency deteriorates
Solution Approach 1:
The patent divides agents into subsets and selects only certain subsets to broadcast in each iteration, maintaining complete model update exchange through coordinated subset selection while reducing total energy consumption by limiting the number of active broadcasters.
Solution Approach 2:
The patent employs dynamic subset selection where the coordinator entity adapts which agents broadcast based on the current state of the learning process and network conditions. This dynamic approach optimizes the balance between model update completeness and energy efficiency by adjusting broadcasting participation in real-time.
Data Source
AI summary
There is provided techniques for selecting agent entities to broadcast local model parameter vectors in an iterative learning process. A method is performed by a coordinator entity. The method, for each iteration of the iterative learning process, comprises obtaining parameters from agent entities. The parameters pertain to a utility for each of the agent entities to broadcast its local model parameter vector for the iteration. The method, for each iteration of the iterative learning process, comprises selecting K<N agent entities to broadcast their local model parameter vector for the iteration by applying a selection criterion to the obtained parameters. The method, for each iteration of the iterative learning process, comprises sending information that informs the N agent entities of the selected K agent entities.