Scienceadvances Miwen copy

Our paper on disaggregated deep learning using in-physics computing directly at RF was accepted to Science Advances. In this paper, we introduced and demonstrated a wireless edge inference architecture that enables disaggregates model access by broadcasting ML model weights over the air and performs complex-valued matrix-vector multiplications directly at the edge devices in the RF domain. This series of work is in collboration with Prof. Dirk Englund’s group at MIT. We will also present this work at the Workshop on AI and ML for Next-Generation Wireless Communications and Networking (AI4NextG) at NeruIPS 2025. [experimental paper] [theoretical paper]