ViT-Reg: Regression-Focused Hardware-Aware Fine-Tuning for ViT on TinyML Platforms

Abstract

Editor’s notes: This article presents a hardware-aware methodology for modeling the performance and energy behavior of vision transformers (ViTs) across a wide design space relevant to TinyML platforms. It serves as a valuable framework for balancing accuracy and efficiency in deploying ViTs on resource-constrained edge systems such as autonomous robotics. —Theocharis Theocharides, University of Cyprus, Cyprus —Marian Verhelst, KU Leuven, Belgium —Vijay Janapa Reddy, Harvard University, USA —Evgeni Gousev, Qualcomm, USA