Many widely used activation functions, including ReLU, GELU, Swish, and Sigmoid, produce predominantly positive outputs, resulting in a nonzero post-activation mean that we term activation bias. Despite its ubiquity, the effect of activation bias on learning dynamics remains poorly understood. Using controlled models,...
Zi-Rui Peng, Helen Zheng, Zi-Ming Liu· 0 citations
This work connects Adam loss spikes to both the mismatch between momentum timescales and finite-scale superquadratic loss geometry beyond the Hessian, and shows that confident cross-entropy losses develop a core--wall landscape comprising a narrow quadratic core followed by a steep wall, which produces effective superq...
It is shown that when a video model generates physically incorrect motion, the correct motion remains available inside the model and can still be made to control the generated video, and a stronger downstream write can restore physical motion, while excessive gain overshoots.
Xing-Yun Wang, Hao-Min Zheng, Man Yuan et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.