AI Paper Atlas

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

Retaining Knowledge for Learning with Dynamic Definition

Zichang Liu, Benjamin Coleman, Tianyi Zhang, Anshumali Shrivastava

NeurIPS 2022 · Continual learning › Continual fine-tuning of LLMs · approach: Continual pretraining of language models · 0 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
replace final layer or model with RIDDLE instead of naive UpdateMLP · MNIST Binary, original test after updateaccuracy: 69.1 -> 98.3Improved
replace EWC with RIDDLEMLP · MNIST Binary, original test after updateaccuracy: 90.9 -> 98.3Improved
replace final layer with RIDDLE instead of Update-ViTViT · CIFAR10, original test after updateaccuracy: 74.4 -> 89.0Improved
replace final layer with RIDDLE instead of EWC-ViTViT · CIFAR10, original test after updateaccuracy: 84.5 -> 89.0Improved
replace baseline with RIDDLE instead of Update-ViTViT · ImageNet, original test after updateaccuracy: 55.6 -> 67.1Improved
replace baseline with RIDDLE instead of Update-RoBERTaRoBERTa · News, original test after updateaccuracy: 74.0 -> 89.7Improved
RIDDLE instead of Update-ViT on update test setViT · ImageNet, update test set after updateaccuracy: 71.5 -> 70.5Hurt
RIDDLE instead of EWC-RoBERTa on update test setRoBERTa · News, update test set after updateaccuracy: 87.5 -> 86.4Hurt
add parameter-count-dependent learning rate to RIDDLE (ablation)RIDDLE-MLP · MNIST Binary, original test after updateaccuracy: 94.9 -> 98.3Improved
use RIDDLE (hash-partitioned parameters) with global learning rate versus Update-MLP (ablation)RIDDLE-MLP · MNIST Binary, original test after updateaccuracy: 69.1 -> 94.9Improved
use pretrained ViT instead of Resnet18 under naive UpdateUpdate · CIFAR10, original test after updateaccuracy: 63.22 -> 74.4Improved
use RIDDLE instead of NN2 on tabular dataRIDDLE · HARaccuracy: 0.9123 -> 0.9001Hurt

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