Machine Learning Model Training via Shared Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limited availability of medical images for training due to patient privacy and data security concerns results in fewer training samples for machine learning models, leading to poor accuracy in medical image processing.
Innovation Solution
A method and apparatus for training machine learning models by acquiring two distinct training sample sets from different hospitals, performing multiple rounds of model training on each set, and sharing model parameters between them to enhance model accuracy without sharing actual images, ensuring data security.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If medical images are shared between hospitals to increase training samples, then model accuracy is improved, but data security and patient privacy are compromised
Solution Approach 1:
The patent creates synthetic training samples by copying and transforming features from existing medical images through a generative model. These synthetic images replicate the statistical characteristics and diagnostic value of real medical images without containing actual patient data, thereby increasing training sample quantity while maintaining data security and patient privacy
Solution Approach 2:
The patent introduces a generative model as an intermediary between real medical images and training samples. This intermediary transforms real images into synthetic images that preserve useful features while removing patient-identifiable information, enabling cross-hospital data utilization without direct data sharing
2Quantity of substance
If more training samples are collected from multiple hospitals, then model accuracy is improved, but data security concerns prevent sharing
Solution Approach 1:
The system generates synthetic training samples by copying features from real medical images through generative models. These synthetic copies maintain the statistical properties and diagnostic information needed for training while being indistinguishable from real images, enabling unlimited sample generation without data security risks
Solution Approach 2:
The patent transforms real medical images into synthetic images by changing parameters such as pixel values, intensities, and spatial characteristics while preserving the underlying diagnostic features. This parameter transformation allows unlimited sample generation from a single hospital's data without creating data security vulnerabilities
3Object-affected harmful factors
If medical images are not shared between hospitals, then data security is maintained, but model training accuracy deteriorates due to limited samples
Solution Approach 1:
The system creates synthetic copies of medical images that preserve diagnostic features while eliminating patient-identifiable information. These synthetic copies can be freely shared and used for training across hospitals without compromising data security, thereby maintaining both security and training effectiveness
Solution Approach 2:
The generative model serves as an intermediary that enables hospitals to share synthetic rather than real images. This intermediary allows training data to be exchanged between hospitals while maintaining data security, as the synthetic images contain no actual patient information
Data Source
AI summary
The present application relates to a method and an apparatus for training machine learning models, a computer device, and a storage medium. The method includes: acquiring a first training sample set and a second training sample set; performing, based on the first training sample set, multiple rounds of model training, to obtain a first machine learning model; and performing, based on the second training sample set, multiple rounds of model training, to obtain a second machine learning model. At least a part of the first machine learning model has a same structure as at least a part of the second machine learning model, at least a part of model parameters of the second machine learning model is used when the first machine learning model is trained, and at least a part of model parameters of the first machine learning model is used when the second machine learning model is trained.