AI Paper Atlas

Research map / Continual learning / Online and replay-based

Repairing Neural Networks by Leaving the Right Past Behind

Ryutaro Tanno, Melanie F. Pradier, Aditya Nori, Yingzhen Li

NeurIPS 2022 · Continual learning › Online and replay-based · approach: Theoretical analysis of forgetting

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 EWC-influence instead of SGD-influence for cause identificationEWC-Influence · CIFAR-10 label noisecomputation time (s): 8301.1 -> 496.3Improved
use EWC-influence instead of SGD-influenceEWC-Influence · MNIST label noisecomputation time (s): 185.1 -> 9.8Improved
use EWC-influence instead of linear influence (SA)EWC-Influence · CIFAR-10 label noisecomputation time (s): 1322.4 -> 496.3Improved
use EWC-influence instead of linear influence (GD)EWC-Influence · MNIST adversarial poisoningcomputation time (s): 10.0 -> 14.5No clear effect
use EWC-influence to rank causes vs other influence methodsEWC-Influence · MNIST/CIFAR-10 label noisefraction of identified causes with wrong labelsImproved
remove top causes by EWC-influence then fine-tune vs other influence methodsEWC-Influence · MNIST/CIFAR-10 label noiseaccuracy on failure setImproved
remove many (about 10^3) identified causes vs remove only annotation-noise instancesall influence methods · MNIST label noisefailure accuracyImproved
remove EWC-identified causes vs remove all injected random input noiseEWC-Influence · MNIST/CIFAR-10 random input noiseaccuracy on failure setImproved
remove identified causes vs randomly removing dataall influence methods · MNIST/CIFAR-10 label noiseaccuracy on failure setImproved
shrink the failure query set to 5% (about 20 examples)all methods · MNIST label noisefailure accuracyNo clear effect
use EWC-deletion instead of Newton update removalEWC-deletion · CIFAR-10 label noisebest failure accuracy: by 5%Improved
use EWC-deletion instead of Newton update removalEWC-deletion · CIFAR-10 label noisebest accuracy on remaining set: less than 1% differenceMixed

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