CT Federated Learning Control Using Client Training Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Federated learning systems face challenges with clients executing inappropriate training, leading to potential contamination and mistraining of global models, especially in the presence of malicious attackers or unintended data usage.
Innovation Solution
Implementing constraints on clients, such as an upper limit on the number of training executions and a time period, to control the training process and reduce the influence of malicious attacks and unintended mistraining, while ensuring privacy through local data usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If federated learning is implemented with multiple clients training local models, then additional training of the global model is achieved without transmitting training data, but some clients may execute training with inappropriate data leading to model contamination
Solution Approach 1:
The system performs preliminary actions by establishing control information (constraints) on clients before training execution. The central server assigns constraints such as maximum training iterations, time limits, or data quantity limits to each client beforehand. This preliminary constraint assignment prevents clients from executing inappropriate training with excessive or malicious data, thereby protecting the global model from contamination while enabling federated learning to proceed.
2Loss of information
If clients are allowed to execute training freely, then local data privacy is protected, but the influence of malicious attacks and unintended mistraining increases
Solution Approach 1:
The system changes parameters by dynamically adjusting training constraints based on client behavior and training logs. The central server monitors training execution and modifies control information (such as reducing time limits or iteration counts) for clients showing signs of inappropriate training. This parameter adjustment maintains privacy protection while controlling training quality and preventing malicious influence.
3Reliability
If constraints are imposed on training execution, then the influence of malicious attacks is reduced, but the complexity of controlling the training process increases
Solution Approach 1:
The system implements feedback mechanisms where clients report training logs back to the central server, which then adjusts control information accordingly. The central server receives training status information from clients and uses this feedback to dynamically modify constraints for subsequent training rounds. This feedback loop enables automated adaptation of control measures without requiring complex manual intervention, balancing reliability improvement with manageable system complexity.
Data Source
AI summary
An X-ray CT apparatus according to an embodiment includes an X-ray tube, an X-ray detector, a processor, and a memory. The memory stores a global model to be used in federated learning. The processor generates CT image data by executing reconstruction processing on detection data of X-rays. The processor transmits, to a client, the global model and control information controlling execution of a trainer at the client. The processor acquires, from the client, a local model generated by training of the global model with training data by the trainer under control of the control information. The processor updates the control information in accordance with a training log of the client. A model generation system according to another embodiment includes a central server and a client. Still another embodiment discloses a model generation method implemented by a client and a central server.