method · first seen 13 Oct 2026
input loss curvature
Also: Hessian norm of loss
Measure of how sharply a model's training loss changes with respect to its inputs, used as a memorization proxy.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Loss Curvature as a Sampling-Free Proxy for Memorization in Diffusion ModelsPapers · 13 Oct 2026 · Shashank Sule, Xinyue Yu, Boonam Shin, Yuxuan Liu, Abolfazl Hashemi, Hayden Scha…
Latest sources
- Loss Curvature as a Sampling-Free Proxy for Memorization in Diffusion ModelsPapers · 13 Oct 2026 · Shashank Sule, Xinyue Yu, Boonam Shin, Yuxuan Liu, Abolfazl Hashemi, Hayden Scha…