Reproduce “ULU: A Unified Activation Function” 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
Simin Huo
ULU uses separate adjustable response curves for negative and nonnegative inputs, retaining a different shape on each side of zero. In a controlled CIFAR-10 classification test with ResNet-18, ULU(0.3, 0.8) achieved 88.7% mean top-1 accuracy over 10 runs, versus 87.9% for Mish and 86.7% for ReLU. This result is specific to the tested setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.