Distributed Learning with Memory: Optimizing Model Usage across Training Tasks

Paper in proceeding, 2024

As the relevance of distributed learning to vehicular services grows, it becomes more important to perform such learning in the most effective possible manner. In this paper, we investigate the benefits stemming from a learning controller considering multiple learning tasks at the same time. Our performance evaluation shows that this new paradigm, which also enables model reusage across learning tasks, yields up to 60% savings on model training costs.