One-Step Curvature Probes Miss the Fitting Operator: Retained Capacity and Terminal Null-Space Correction for Continual Learning
A one-step curvature probe evaluates an initial direction, whereas continual learners are judged after reaching comparable new-task fit. In an overparameterized linearization, projected gradient descent converges to $\Delta_P=PJ^\top(JPJ^\top)^{-1}r$, and its squared-displacement inflation is…
By Abu Sa-Adat Mohamed Moon-Im Al Ahsan, Ibne Farabi Shihab, Md Najmus SwaqeebarXiv cs.LG
Read the full story
Originally published by arXiv cs.LG on Oct 6, 2026.
Comments
Sign in to join the discussion.