AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Better Fine-Tuning by Reducing Representational Collapse

Armen Aghajanyan, Akshat Shrivastava, Anchit Gupta, Naman Goyal, Luke Zettlemoyer, Sonal Gupta

ICLR 2021 · Architectures and efficiency › Pruning and efficient tuning · approach: Adapter-based parameter-efficient tuning · 20 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 R3F regularization during RoBERTa-Large fine-tuningRoBERTa-Large · GLUE development taskstask metrics (accuracy or Matthews correlation)Improved
Replace adversarial SMART regularization with parametric-noise R3F regularizationRoBERTa-Large · GLUE development taskstask metrics (accuracy or Matthews correlation)Mixed
Add R3F regularization to XLM-R Large fine-tuningXLM-R Large · XNLI test zero-shotaverage accuracy across languages: 80.9 -> 81.2Improved
Add R4F regularization, including spectral normalization of the classification headXLM-R Large · XNLI test zero-shotaverage accuracy across languages: 80.9 -> 81.4Improved
Add R3F regularization to BART fine-tuningBART · CNN/DailyMailROUGE-1: 44.16 -> 44.38Improved
Add R3F regularization to BART fine-tuningBART · GigawordROUGE-1: 39.29 -> 40.45Improved
Add R3F regularization to BART fine-tuningBART · Reddit TIFU LongROUGE-1: 24.19 -> 30.31Improved
Use trust-region-style regularization instead of standard fine-tuningRoBERTa fine-tuned on SST-2, then probed on other GLUE tasks · GLUE probing experimentsprobe performanceImproved
Use R3F or R4F rather than standard fine-tuning through sequential task updatesRoBERTa · Sequential GLUE probingsource-task probe performanceImproved

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