Inference and Latency in Mobile Edge Systems

How should we manage on-device ML/AI with edge assitance?

Deploying complex machine learning models on resource-constrained devices is challenging due to limited computational power, memory, and model retrainabil- ity. One proposal is to have mobile devices offload computation to processing centers which are physically closer than the abstract “cloud.” These mobile edge cloud (MEC) systems raise a number of interesting challenges in terms of latency, accuracy, energy efficiency, and privacy (among others).

Representative publications

  • Yu Wu, Yansong Li, Zeyu Dong, Nitya Sathyavageeswaran, and Anand D. Sarwate, Learning to Help in Multi-Class Settings, In The Thirteenth International Conference on Learning Representations (ICLR 2025), Apr 2025.

Support

  • NSF CNS-2148104 RINGS: REALTIME: Resilient Edge-cloud Autonomous Learning with Timely Inferences (PI: Anand D. Sarwate, Co-PIs: Waheed U. Bajwa, Dipankar Raychaudhuri, Roy D. Yates)