In-Network Aggregation: Faster Distributed ML Training Distributed machine learning can be limited by the exchange of model updates, not just by compute speed. SwitchML explores in-network aggregation, using programmable switches to combine updates and reduce communication overhead during training. Read More →