Research combines privacy tech with active learning for oncology AI workflows

A developer has created a prototype system designed for precision oncology clinical workflows. It uses privacy-preserving federated learning to train AI models on sensitive patient data without centralizing it. The system incorporates active learning to efficiently select data for labeling across different hospital sites. A novel verification method called inverse simulation allows clinicians to audit model behavior. This approach addresses legal and statistical challenges in training AI on rare, sensitive medical data.