What AI says about Ensembled Membership Auditing (EMA)

EMA (Ensembled Membership Auditing) provides code and workflows to audit data removal from trained models and assess membership inference risks. The repository includes network architectures, data utilities, training scripts for base and calibration models, an auditing runner, and a script to aggregate results into a final score for evaluation. It supports experiments on datasets like MNIST, COVIDx, and Child X-ray to demonstrate the EMA methodology described in the paper EMA: Auditing Data Removal from Trained Models.