AI Paper Atlas

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

Beyond Majority Voting: LLM Aggregation by Leveraging Higher-Order Information

Rui Ai, Yuqi Pan, David Simchi-Levi, Milind Tambe, Haifeng Xu

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression · 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 ISP instead of majority votingLLM aggregation · Simulated (K=4)accuracy: 92.64% -> 94.45%Improved
use ISP instead of majority votingLLM aggregation · Simulated (K=2)accuracy: 85.13% -> 90.48%Improved
use ISP instead of majority votingLLM aggregation · Simulated (K=10)accuracy: 95.54% -> 96.49%Improved
use surprisingly popular instead of majority votingLLM aggregation · Simulated (K=2)accuracy: 85.13% -> 79.94%Hurt
use OW-L instead of majority votingfour strong LLMs · UltraFeedbackaccuracy: 72.21% -> 73.66%Improved
use OW-L instead of majority votingfour strong LLMs · MMLUaccuracy: 89.32% -> 90.37%Improved
use OW-L instead of majority votingfour strong LLMs · ARMMANaccuracy: 85.24% -> 85.78%Improved
aggregate strong models with OW instead of picking oracle single bestfour strong LLMs · MMLUaccuracy: 91.02% -> 90.37%Hurt
use OW-L instead of majority voting on hard-distractor subsetfour strong LLMs · MMLU-hardaccuracy: 17.23% -> 24.79%Improved
use ISP instead of majority voting on hard-distractor subsetfour strong LLMs · MMLU-hardaccuracy: 17.23% -> 18.66%Improved
ensemble all eight models with OW-I instead of MV (ablation)eight LLMs · UltraFeedbackaccuracy: 70.23% -> 72.44%Improved
ensemble all eight models with OW-L instead of MV (ablation)eight LLMs · ARMMANaccuracy: 84.40% -> 85.10%Improved

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