AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Batch and Sequential Unlearning for Neural Networks

Nhung Bui, Xinyang Lu, Rachael Hwee Ling Sim, See-Kiong Ng, Bryan Kian Hsiang Low

ICLR 2026 · Learning theory and algorithms › Prediction and decision losses · approach: Data attribution and recourse

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use pseudo-inverse of the Hessian instead of exact inverse (PINV-Newton)Newton unlearning · CNN FMNIST class-level batchToW: 0.13 -> 0.05Hurt
add cubic regularization with automatically chosen damping (CuReNU) to Newton unlearningDamped Newton · CNN FMNIST class-level batchToW: 0.05 -> 0.93Improved
use CuReNUS rather than CuReNUCuReNU · CNN FMNIST class-level batchToW: 0.93 -> 0.99Improved
use CuReNU/CuReNUS update instead of damped NewtonDamped Newton · CNN FMNIST class-level batchNewton update norm: 838.68 ± 742.96 -> 0.38 ± 0.05Improved
use second-order CuReNUS instead of first-order GDGD · CNN FMNIST class-level batchToW: 0.14 -> 0.99Improved
use CuReNUS instead of GD in sequential unlearningGD · Llama-2 TOFU sequential (last round)ToW: 0.60 -> 0.80Improved
use CuReNUS instead of SCRUB in sequential unlearningSCRUB · Llama-2 TOFU sequential (last round)ToW: 0.72 -> 0.80Improved
use CuReNUS instead of SCRUB in sequential unlearningSCRUB · ResNet-18 CIFAR-10 sequential (last round)ToW: 0.944 -> 0.909No clear effect
use CuReNUS instead of GA/Rand. Lbls. in sequential unlearningGA · ResNet-18 CIFAR-10 sequential (last round)ToW: 0.143 -> 0.909Improved
use CuReNUS (stochastic Hessian-free) instead of SCRUB for efficiencySCRUB · Llama-2 AG-News unlearning timeunlearning time (s): 6796.16 -> 85.26Improved
use CuReNUS (Hessian-free) instead of CuReNU for efficiencyCuReNU · CNN FMNIST unlearning timeunlearning time (s): 6355.31 -> 35.54Improved
use CuReNUS instead of CuReNU, peak memoryCuReNU · CNN FMNIST peak memorypeak memory (MB): 6226 -> 1588Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy