Session

DescriptionWhile modern high-performance computing systems enable and seamlessly infuse AI/ML into our work and daily lives, AI/ML can also be applied to improve the operational efficiency and robustness in computing reciprocally. Optimization occurs in both directions to continuously elevate the quality of AI services and the performance of systems running AI, creating a positive cycle. This session presents such a category of techniques, including memory optimization for DNN on tiny devices, enhancement for adversarial robustness, an adaptive federated learning scheme, adaptive multi-tensor fusion, optimization for edge ML, and an auto-tuning framework for real-time vision tasks.