AI Paper Atlas

Research map / Large language models / Alignment and preferences

Preference-based Antibody Expression Ranking: Scaling with Large-scale Weak Supervision

Josh Qixuan Sun, Morteza Babaie, Wenyang hou, Mark Crowley, David Young

ICML 2026 · Large language models › Alignment and preferences · approach: DPO design choices · 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
add Stage I continual pretraining before preference optimization (ablation)Ours (AntiBERTa2) · Exp-1K test MCCMCC: 26.7 -> 43.7Improved
add Stage I continual pretraining before preference optimization (ablation)Ours (AntiBERTa2) · Exp-1K test TauKendall Tau: 33.8 -> 46.3Improved
add Stage I continual pretraining before supervised regression (ablation)Supervised (AntiBERTa2) · Exp-1K test composite AvgAvg: 33.1 -> 29.0Hurt
add Stage I continual pretraining before supervised regression (ablation)Supervised (IgBERT) · Exp-1K test composite AvgAvg: 17.2 -> 37.5Improved
replace supervised regression with preference optimization on same Stage I backboneAntiBERTa2 · Exp-1K test composite AvgAvg: 29.0 -> 45.0Improved
add Camel-4M weak supervision to DPO (ablation)Ours (AntiBERTa2) · Exp-1K MCCMCC: 32.8 -> 43.7Improved
add Camel-4M weak supervision to DPO (ablation)Ours (AntiBERTa2) · Exp-1K TauKendall Tau: 35.1 -> 46.3Improved
use DPO instead of regression on labeled data only (no Camel-4M)AntiBERTa2 · Exp-1K TauKendall Tau: 37.6 -> 35.1Hurt
raise sampling rate alpha from 10 to 1000 (ablation)Ours · Exp-1K TauKendall Tau: 12.2 -> 40.5Improved
raise sampling rate alpha from 1000 to 10000 (ablation)Ours · Exp-1K TauKendall Tau: 40.5 -> 30.7Hurt
filter preference pairs by similarity threshold 0.59 (ablation)Ours (AntiBERTa2) · Exp-1K MCCMCC: 43.7 -> 6.4Hurt
add weak data to reduce generalization gap (ablation)Ours · |Avg val - Avg test|delta Avg: 14.1 -> 5.5Improved

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