Research map / Large language models / Retrieval and knowledge
Embracing Positional Bias in Multiple-Choice Question Answering via Permutation Equivariant Neural Networks
Chengyu Jiao, Siyin Huang, Yu Zhang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add EMBER permutation-equivariant network on top of the LLM | Llama-3.1-8B · MMLU | accuracy: 52.2 -> 57.4 | Improved |
| add EMBER | Falcon-7B · MMLU | accuracy: 24.8 -> 33.2 | Improved |
| add EMBER instead of PriDe or UniBias | Qwen-2.5-7B · MMLU | accuracy: 55.5 (PriDe) -> 56.9 | Improved |
| add EMBER | Llama-3.2-3B · ARC-Challenge | accuracy: 40.2 -> 46.5 | Improved |
| add EMBER under permutation attack (any wrong) | Llama-2-7B · MMLU STEM any wrong | any-wrong accuracy: 6.4 -> 36.0 | Improved |
| replace cross attention with subtraction as difference function (ablation) | EMBER · MMLU (Llama-2-7B) | accuracy: 39.49 -> 36.69 | Hurt |
| add divergence loss (ablation) | EMBER · MMLU (Llama-2-7B) | any-wrong accuracy: 12.92 -> 39.38 | Improved |
| add divergence loss (ablation) | EMBER · MMLU (Llama-2-7B) | accuracy: 39.14 -> 39.49 | Improved |
| vary decay factor from 0.9 to 0.1 (ablation) | EMBER · MMLU (Llama-2-7B) | accuracy gain: 6.42 -> 6.36 | No clear effect |
| Existing debiasing (PriDe, UniBias) versus equivariance under permutation attack | Llama-3.1-8B · MMLU STEM any wrong | drop from regular accuracy: PriDe -4.8, UniBias -7.2 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- UniBias — UniBias: Unveiling and Mitigating LLM Bias through Internal Attention and FFN Manipulation (NeurIPS 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Large Language Models Are Not Robust Multiple Choice Selectors (ICLR 2024 · similarity 0.88)
- Set-LLM: A Permutation-Invariant LLM (NeurIPS 2025 · similarity 0.85)
- Fool Your (Vision and) Language Model with Embarrassingly Simple Permutations (ICML 2024 · similarity 0.85)
- Listening to the Wise Few: Query–Key Alignment Unlocks Latent Correct Answers in Large Language Models (NeurIPS 2026 · similarity 0.84)
- Eliminating Position Bias of Language Models: A Mechanistic Approach (ICLR 2025 · similarity 0.84)
- Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions (ICLR 2025 · similarity 0.83)
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.
