Reproduce “Empowering On-Device Model Adaptation with an Edge AI Inference Accelerator” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mateusz Piechocki, Alessandro Capotondi, Marek Kraft
The authors present an on-device adaptation method that runs a frozen INT8 backbone on a Hailo-8L and trains an FP32 classification head on the host CPU. This approach reaches up to 15.4× faster training than the Raspberry Pi 5 CPU for ResNet18, with 64.62% accuracy on batch-one CIFAR-100 after five epochs. It evaluates image-classification adaptation, not full-network training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.