Research map / Vision-language and multimodal / Multimodal LLMs
Calibrating MLLM-as-a-judge via Multimodal Bayesian Prompt Ensembles
Eric Slyman, Mehrab Tanjim, Kushal Kafle, Stefan Lee
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 image-cluster-conditioned prompt weights (MMB vs BPE) | Bayesian prompt ensemble · HPSv2 pairwise (10 prompts, 20 validation samples) | NLL: .547 -> .430 | Improved |
| add image-cluster-conditioned prompt weights (MMB vs BPE) | Bayesian prompt ensemble · HPSv2 pairwise (10 prompts, 20 validation samples) | F1: .753 -> .778 | Improved |
| add image-cluster-conditioned prompt weights (MMB vs BPE) | Bayesian prompt ensemble · HPSv2 pairwise (10 prompts, 20 validation samples) | Brier score: .147 -> .135 | Improved |
| add image-cluster-conditioned prompt weights (MMB vs BPE) | Bayesian prompt ensemble · HPSv2 pairwise (10 prompts, 20 validation samples) | ECE: .114 -> .091 | Improved |
| use MMB instead of unweighted averaging | prompt ensemble · HPSv2 pairwise (10 prompts, 20 validation samples) | F1: .764 -> .778 | Improved |
| use MMB instead of unweighted averaging | prompt ensemble · HPSv2 pairwise (10 prompts, 20 validation samples) | ECE: .142 -> .091 | Improved |
| use ensemble (BPE vs Avg.) weighting | prompt ensemble · HPSv2 pairwise (10 prompts, 20 validation samples) | F1: .764 -> .753 | Hurt |
| use MMB instead of the best single prompt | single prompt · HPSv2 pairwise (10 prompts, 20 validation samples) | accuracy: .812 -> .820 | Improved |
| increase cluster count K from 4 to 16 (ablation) | MMB · HPSv2 pairwise (10 prompts, 20 validation samples) | NLL: .432 -> .430 | Improved |
| increase cluster count K from 4 to 64 (ablation) | MMB · HPSv2 pairwise (10 prompts, 20 validation samples) | ECE: .090 -> .091 | No clear effect |
| use MMB instead of unweighted averaging | prompt ensemble · MJBench-Bias synthetic no-preference | average confidence: .656 -> .705 | Hurt |
| use MMB instead of BPE | prompt ensemble · MJBench-Bias synthetic no-preference | average confidence: .715 -> .705 | Improved |
Closest papers by meaning
- MJ-Bench: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation? (NeurIPS 2025 · similarity 0.90)
- Multimodal Prompt Optimization: Why Not Leverage Multiple Modalities for MLLMs (ICLR 2026 · similarity 0.89)
- DynT2I-Eval: A Dynamic Evaluation Framework for Text-to-Image Models (NeurIPS 2026 · similarity 0.88)
- MLLM-as-a-Judge: Assessing Multimodal LLM-as-a-Judge with Vision-Language Benchmark (ICML 2024 · similarity 0.88)
- SEED-Bench: Benchmarking Multimodal Large Language Models (CVPR 2024 · similarity 0.88)
- LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models (ICLR 2026 · similarity 0.88)
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.
