dsMTL

dsMTL provides a computational framework for federated multi-task learning, enabling simultaneous learning of related regression or classification tasks by exploiting shared information across distributed data sources while preserving privacy.

Key Features:

• Federated Multi-Task Learning: Supports analysis of sensitive, individual-level data from geographically distributed data sources using the DataSHIELD platform.
• Differential Privacy: Integrates differential privacy using the Laplace mechanism as an optional feature to obscure individual-level characteristics from the model while retaining group-level differences.
• Privacy-Performance Balance: Enables selection of a privacy parameter to balance the degree of privacy protection with conservation of model performance.

Scientific Applications:

• Distributed Sensitive Data Analysis: Enables multi-task learning on sensitive data that cannot be centralized due to privacy or regulatory constraints.
• Membership Inference Attack Protection: Protects against attacks attempting to determine whether a specific individual was included in a given training set using the model.
• Disease Classification from Gene Expression: Identifies patient groups such as schizophrenia patients from distributed gene expression data.

Methodology:

The Laplace mechanism is used to implement differential privacy, with validation performed through simulation studies and a case study using gene expression data for schizophrenia patient identification.