Related Experiment Video
Updated: Sep 27, 2025
07:13
Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities
Published on: October 27, 2023
1.4K
Handling data heterogeneity with generative replay in collaborative learning for medical imaging
Liangqiong Qu1, Niranjan Balachandar1, Miao Zhang1
1Department of Biomedical Data Science at Stanford University, Stanford, CA 94305, USA.
Medical Image Analysis
|April 7, 2022
Summary
This study introduces a new generative replay method to improve privacy-preserving deep learning in healthcare. The technique effectively handles varied data across institutions, boosting model accuracy.
Area of Science:
- Artificial Intelligence
- Machine Learning
- Medical Informatics
Background:
- Collaborative learning enables decentralized, privacy-preserving training of deep neural networks across multiple institutions, particularly valuable in healthcare.
- A significant challenge in collaborative learning is data heterogeneity, where data distributions vary considerably between institutions.
- Existing methods often aggregate model parameters directly, which can be suboptimal with heterogeneous data.
Purpose of the Study:
- To address the challenge of data heterogeneity in collaborative learning for healthcare applications.
- To propose a novel generative replay strategy that aggregates knowledge from heterogenous clients in a privacy-preserving manner.
- To improve the performance of collaborative deep learning models trained on diverse datasets.
Main Methods:
- Developed a novel dual model architecture comprising a primary model for task performance and an auxiliary generative replay model.
- Leveraged generative adversarial learning to aggregate knowledge from heterogenous clients, rather than direct parameter aggregation.
- The auxiliary model broadcasts aggregated knowledge to the central server to regulate the primary model's training with an unbiased target distribution.
Main Results:
- The proposed generative replay strategy effectively handles heterogeneous data across institutions.
- Achieved approximately 4.88% improvement in prediction accuracy for diabetic retinopathy classification on highly heterogeneous data.
- Demonstrated a reduction of approximately 49.8% in mean absolute value for Bone Age prediction on highly heterogeneous data.
Conclusions:
- The novel generative replay strategy significantly enhances collaborative learning performance in the presence of data heterogeneity.
- This approach offers a more robust method for training deep neural networks across multiple institutions with diverse datasets.
- The findings highlight the potential of generative replay for advancing privacy-preserving, decentralized AI in healthcare.
Related Concept Videos
02:07Improving Translational Accuracy
11.9K
Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
11.9K
01:12Observational Learning
348
Albert Bandura's observational learning, also known as imitation or modeling, occurs when a person observes and imitates another's behavior. It is a quicker process than operant conditioning. A well-known example is the Bobo doll study, where children who saw an adult acting aggressively towards the doll were more likely to act aggressively when left alone, compared to those who observed a nonaggressive adult. Many psychologists view observational learning as a form of latent learning...
348