A recent work from CRCV researchers and collaborators was a finalist for the Best Paper Award at the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2022. The paper is titled Local Learning Matters: Rethinking Data Heterogeneity in Federated Learning. In a nutshell, the work investigates the data heterogeneity challenge in Federated Learning (FL). Essentially, this refers to the difficulty in training a model from a distributed network of clients with varying local data distributions in a privacy-preserving manner. The work approaches this challenge from a unique angle compared to other previous works. Simply put, it emphasizes that local learning is extremely important in federated settings, and finds that methods focused on promoting learning generality inherently improve global FL aggregation and optimization to a surprising degree. For more details, check out the paper and code.