AI Paper Atlas

Research map / Continual learning / Continual fine-tuning of LLMs

Learning with Recoverable Forgetting

Jingwen Ye, Yifang Fu, Jie Song, Xingyi Yang, Songhua Liu, Xin Jin, Mingli Song, Xinchao Wang

ECCV 2022 · Continual learning › Continual fine-tuning of LLMs · approach: Machine unlearning by selective forgetting · 32 citations (OpenAlex)

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
add pruning-based initialisation of the deposit module (ablation)LIRF · CIFAR-10Avg Acc (recover net): 95.15 -> 95.49Improved
add attention loss Lat to the knowledge-preservation term (ablation)LIRF · CIFAR-10Avg Acc (recover net): 95.09 -> 95.15No clear effect
add attention loss Lat to Lkr (ablation)LIRF target net · CIFAR-10, ResNet18 student, AT-based distillationDep Acc of distilled student: 94.61 -> 88.76Improved
add attention loss Lat to Lkr (ablation)LIRF target net · CIFAR-10, CNN student, AT-based distillationDep Acc of distilled student: 85.72 -> 81.83Improved
add attention loss Lat to Lkr (ablation)LIRF target net · CIFAR-10, ResNet18 student, logit-based distillationDep Acc of distilled student: 92.70 -> 91.49Improved
add attention loss Lat to Lkr (ablation)LIRF target net · CIFAR-10, CNN student, logit-based distillationDep Acc of distilled student: 84.26 -> 82.75Improved
add Lpt and Lre to Lkr+Lkp (w/o Lat) (ablation)LIRF · CIFAR-10Avg Acc (recover net): - -> 95.09Improved
replace incremental-learning-style training with LIRF deposit/withdrawrecover net · CIFAR-10Avg Acc: 93.12 -> 95.49Improved
use LIRF recover net instead of original networkResNet-18 · CIFAR-10Avg Acc: 94.06 -> 95.57Improved
use LIRF recover net instead of original networkResNet-18 · CUB200-2011Avg Acc: 49.81 -> 54.01Improved
increase division block number n (ablation)LIRF · CIFAR-10Dep Acc and Avg AccHurt
increase pruning rate towards 50% (ablation)LIRF deposit module · CIFAR-10Avg Acc (recover net)Mixed

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