## Multi-Source Transfer Learning: A New Approach Transfer learning is a fundamental technique for improving model performance, especially in data-scarce scenarios. A new study introduces a theoretical framework called Unified Optimization of Weights and Quantities (UOWQ) to address the challenges of multi-source transfer learning. ## Joint Optimization of Weights and Quantities UOWQ formulates multi-source transfer learning as a parameter estimation problem, based on an asymptotic analysis of a Kullback-Leibler divergence-based generalization error measure. The framework jointly determines the optimal source weights and optimal transfer quantities for each source task. ## Theoretical Results and Practical Algorithms The analysis demonstrates that using all available source samples is always optimal, once the weights have been adjusted correctly. The study also provides closed-form solutions for the single-source case and develops a convex optimization-based numerical procedure for the multi-source case. Based on the theoretical results, practical algorithms have been developed for multi-source transfer learning and multi-task learning. ## Empirical Validation Experiments on real-world datasets, such as DomainNet and Office-Home, demonstrate that UOWQ consistently outperforms strong baselines, validating both the theoretical predictions and the practical effectiveness of the framework.