AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

RaLEs: a Benchmark for Radiology Language Evaluations

Juan Manuel Zambrano Chaves, Nandita Bhaskhar, Maayane Attias, Jean-Benoit Delbrouck, Daniel Rubin, Andreas Markus Loening, Curtis Langlotz, Akshay S Chaudhari

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: Data and evaluation methodology · 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
use 10% of training labels instead of full (ablation)BERT-base · RaLEs NLUaverage score: 81.8 -> 64.2Hurt
domain-specific clinical/radiology pretraining vs general Englishencoder models · RaLEs NLUaverage scoreImproved
replaced-token-detection objective (ELECTRA) vs masked LM onlyELECTRA-base · RaLEs NLUaverage score: 82.0Improved
add document-link prediction objective (BioLinkBERT vs PubMedBERT)PubMedBERT-base · RaLEs NLUaverage score: 81.9 -> 83.3Improved
increase parameters from ~100M to ~300MBERT/RoBERTa/DeBERTa/BioLinkBERT · RaLEs NLUaverage score: 83.0 -> 82.7No clear effect
larger BERT (110M to 340M)BERT · RaLEs NLUaverage score: 81.8 -> 81.4No clear effect
gains on GLUE or BLURB as predictor of radiology NLUgeneral and biomedical models · RaLEs NLUaverage score: 10 percentage point improvements in GLUE or BLURB lead to a 1-2 percent improvementNo clear effect
radiology domain adaptation (RadBERT) on other benchmarksRadBERT1,2 · BLURBscoreHurt
fine-tune with frozen LM weights (ablation)BERT-base · RaLEs NLUaverage score: 81.8 -> 69.4Hurt
clinical GatorTron vs general ELECTRA-base for summarizationELECTRA-base · MEDIQA 2021ROUGE-2: .238 -> .250Improved
clinical GatorTron vs ELECTRA-base for factualityELECTRA-base · BioNLP 2023RadGraph F1: .229 -> .283Improved
out-of-distribution test (Stanford) instead of in-distribution (Indiana)GatorTron · summarizationROUGE-2: .262 -> .238Hurt

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