AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Bi-directional Bias Attribution: Debiasing Large Language Models without Modifying Prompts

Yujie Lin, Kunquan Li, YiXuan Liao, Xiaoxin Chen, Jinsong Su

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression · 1 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 FBA/BBA neuron interventionLlama-3.1-8B · StereoSet GenderICAT: 45.31 -> 62.50Improved
add FBA interventionLlama-3.1-8B · StereoSet ReligionICAT: 78.48 -> 91.14Improved
add FBA/BBA intervention vs prompt methodsLlama-3.1-8B · BBQ ambiguous vs disambiguatedAcc_dis: 73.02 (Self-Debiasing) vs 88.87 (FBA)Improved
add BBA interventionLlama-3.1-8B · WinoBiasGap: 25.26 -> 1.02Improved
add FBA interventionLlama-3.2-3B · WinoBiasGap: 91.42 -> 18.68Improved
add BBA interventionLlama-3.1-8B · Bias-in-BiosGap-TPR: 6.05 -> 1.39Improved
add FBA interventionLlama-3.1-8B · Bias-NLIP_neutral: 81.8 -> 99.1Improved
apply IG2 neuron suppression insteadLlama-3.2-3B · StereoSet GenderLMS: 97.66 -> 66.41Hurt
replace stereotype cue selection with first-word-of-group cues (ablation)FBA, Llama-3.1 · StereoSetSSHurt
replace attribution with random neuron selection (ablation)FBA, Llama-3.1 · StereoSetSS/LMSHurt
intervene on first-layer instead of last-layer neurons (ablation)FBA, Llama-3.1 · StereoSet GenderICAT: 62.50 -> 48.44Hurt
intervene on first-layer neurons (nationality) (ablation)FBA, Llama-3.1 · StereoSet NationalityICAT: 68.19 -> 68.19No clear effect

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