Federated Learning Accuracy via Client Clustering and Exclusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In federated learning for medical AI, there is no effective way to prevent accuracy decreases in the master model due to incorrect data from participating clients, which can lead to misdiagnoses and reduced learning accuracy, especially when multiple medical institutions are involved.
Innovation Solution
A machine learning system that synchronizes learning models across client terminals, creates client clusters, evaluates the inference accuracy of master model candidates, and identifies and excludes clients causing accuracy deterioration, thereby ensuring learning accuracy without extracting sensitive data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If federated learning is used to train AI models without extracting data from medical institutions, then privacy protection is improved, but learning accuracy may deteriorate due to incorrect data from participating clients
Solution Approach 1:
The patent segments clients into multiple clusters and creates multiple master model candidates from different clusters. By dividing the client population into segments and creating separate model candidates for each segment, the system can identify which segments contain incorrect data and isolate their impact, thereby maintaining learning accuracy while preserving privacy through federated learning
Solution Approach 2:
The patent introduces a feedback mechanism where each master model candidate is evaluated against verification data, and clients are identified and excluded based on performance feedback. This feedback loop allows the system to detect and correct the influence of clients with incorrect data, ensuring that only reliable clients contribute to the final master model while maintaining the privacy benefits of federated learning
2Adaptability or versatility
If multiple medical institutions participate in federated learning, then model versatility is improved, but the risk of accuracy deterioration from incorrect data increases
Solution Approach 1:
The patent divides participating medical institutions into multiple clusters and creates separate master model candidates for each cluster. This segmentation allows the system to maintain versatility by incorporating data from multiple institutions while managing the risk of incorrect data by isolating problematic institutions within their clusters and preventing them from contaminating the overall model
Solution Approach 2:
The patent introduces an intermediary evaluation process using verification data that mediates between the diverse data from multiple medical institutions and the final master model. This intermediary step filters out incorrect data while preserving the versatility benefits of multi-institutional participation, as the verification process acts as a gatekeeper before data influences the master model
3Device complexity
If clients with incorrect data are allowed to participate in learning, then system simplicity is maintained, but master model accuracy decreases
Solution Approach 1:
The patent performs preliminary actions by creating multiple master model candidates from different client clusters before final model selection. This preliminary creation of alternative models allows the system to evaluate and compare performance before committing to a final model, enabling the identification and exclusion of clients with incorrect data without requiring complex real-time monitoring or intervention during the learning process
Solution Approach 2:
The patent creates multiple copies of the master model (master model candidates) from different client clusters. By copying the model multiple times with different data subsets, the system can compare performances and identify which copies are affected by incorrect data, allowing for the selection of accurate models while maintaining relative system simplicity through straightforward copying and evaluation procedures
Data Source
AI summary
There are provided a machine learning system and method, an integration server, an information processing apparatus, a non-transitory computer readable medium storing a program, and an inference model creation method capable of ensuring a learning accuracy in federated learning. Each client terminal executes machine learning of the learning model using data stored in a medical institution, and transmits a learning result to the integration server. The integration server divides the plurality of client terminals into a plurality of client clusters, and creates master model candidates by integrating the learning results for each client cluster. The integration server evaluates an inference accuracy of each master model candidate, and in a case where a master model candidate having an inference accuracy lower than an accuracy threshold value is detected, extracts a client terminal as an accuracy deterioration cause from the client cluster used for creation of the master model candidate.