Mobile Body Selection for Federated Learning Efficiency

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In distributed machine learning environments with heterogeneous computing nodes, existing systems face challenges in efficiently selecting suitable mobile bodies for machine learning tasks, often resulting in redundant data collection and processing due to nodes traveling in similar directions, which hampers efficient federated learning.

Innovation Solution

An information processing apparatus and method that acquire and utilize mobile body information, including position and travel direction data, to selectively choose nodes for machine learning, thereby avoiding redundant data collection and enhancing the efficiency of federated learning by selecting nodes traveling in different directions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If distributed machine learning is performed using heterogeneous computing nodes, then machine learning can be performed across multiple mobile bodies, but redundant data collection occurs when nodes travel in similar directions

Engineering Contradiction:
Improvemachine learning efficiencyVSAvoidredundant data collection
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The system performs preliminary acquisition of position information and travel direction information for each mobile body before selecting nodes for machine learning. This preliminary action enables the server to predict potential redundancy in data collection and proactively select nodes that will provide diverse, non-redundant data, thereby improving machine learning efficiency while minimizing redundant data collection.

Inventive Principle:
Principle #10Preliminary action

2Quantity of substance

If nodes traveling in similar directions are selected for machine learning, then more computing nodes can participate, but the diversity of collected data decreases

Engineering Contradiction:
Improvenumber of participating nodesVSAvoiddata diversity
Core Design Contradiction:
Quantity of substanceVSAdaptability or versatility

Solution Approach 1:

The system applies local quality by selecting computing nodes based on their specific local characteristics - namely their travel directions and positions. Instead of treating all nodes uniformly, the server evaluates each node's unique trajectory and selects a subset where each selected node contributes data from a distinct local perspective (different travel direction), thereby maintaining data diversity while maximizing the number of participating nodes.

Inventive Principle:
Principle #3Local quality

3Reliability

If position information is acquired for all mobile bodies, then suitable nodes can be selected, but information processing complexity increases

Engineering Contradiction:
Improvenode selection accuracyVSAvoidinformation processing complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system extracts only the essential information needed for node selection - specifically position information and travel direction information - from the complete set of mobile body data. By taking out only these critical parameters rather than processing all available information about each mobile body, the system achieves reliable node selection based on diversity while minimizing information processing complexity.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4220494A1Information processing apparatus, information processing method, and non-transitory storage medium
Publication Date: 2023.08.02 TOYOTA JIDOSHA KK
  • EP4220494A1 patent drawingFigure 1
  • EP4220494A1 patent drawingFigure 2A~2B
  • EP4220494A1 patent drawingFigure 3

AI summary

An information processing apparatus (3) includes a controller configured to: acquire mobile body information including information that indicates the respective positions of a plurality of mobile bodies (10) that are usable for machine learning; and select two or more mobile bodies (10) to be used for machine learning based on the mobile body information.