This article discusses a novel approach to multi-task learning (MTL), particularly in the context of medical image analysis. The researchers propose formulating MTL as a multi/bi-level optimization problem, forcing features to learn from each task in a cooperative manner. To address the negative transfer problem during optimization, they search for flat minima for the current objective function with regard to features from other tasks. The effectiveness of this method is demonstrated through tests on three publicly available datasets, showing promising results compared to other MTL methods.
Publication date: 21 Sep 2023
Project Page: https://arxiv.org/abs/2309.12090v1
Paper: https://arxiv.org/pdf/2309.12090
