Reproduce “NACHOS: Neural Architecture Search for Hardware Constrained Early Exit Neural Networks” 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
Matteo Gambella, Jary Pomponi, Simone Scardapane, Manuel Roveri
NACHOS searches jointly for an image-classification backbone and its early-exit classifiers under accuracy and inference-compute constraints. It accounts for each classifier’s cost before training and uses confidence-based exits to reduce multiply-accumulate operations (MACs). Across five CIFAR-10 search seeds, NACHOS reaches 72.65 ± 0.49% accuracy at 2.44 ± 0.02 million MACs, versus EDANAS’s 67.78 ± 0.12% at 2.47 ± 0.01 million; these benchmark results measure MACs, not latency or energy, and NACHOS performs worse than expert-designed networks on SVHN.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.