Research map / Learning theory and algorithms / Fairness
Fair in Mind, Fair in Action? A Synchronous Benchmark for Understanding and Generation in UMLLMs
Yiran Zhao, Lu Zhou, Xiaogang Xu, Zhe Liu, Jiafei Wu, Liming Fang
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 |
|---|---|---|---|
| compare UMLLMs against specialist text-to-image models on generation fairness | UMLLMs · IRIS Generation IFS | IFS: Bagel 82.58 vs SD 3.5 Large 273.17 | Hurt |
| evaluate understanding vs generation within same UMLLM | UMLLMs · IRIS Understanding vs Generation | fairness scores: VILA-U: HAF in understanding, UDF in generation | Mixed |
| probe BLIP3-o AR-to-diffusion projection layer | BLIP3-o · mechanistic probes | bias distortion: Distortion ≈ 1.4; Consistency ≈ 0.97 | Hurt |
| use counter-stereotype prompts | UMLLMs · image generation | output quality and semantic fidelity | Improved |
| compare hybrid vs purely autoregressive architectures | UMLLMs · IRIS scores | fairness score: Welch t-test p = 0.76 | No clear effect |
| vary scoring hyperparameters by 10% (ablation) | IRIS scoring · model rankings | Spearman rank correlation: rho > 0.96 | No clear effect |
Closest papers by meaning
- FairMT-Bench: Benchmarking Fairness for Multi-turn Dialogue in Conversational LLMs (ICLR 2025 · similarity 0.87)
- Is Your Model Fairly Certain? Uncertainty-Aware Fairness Evaluation for LLMs (ICML 2025 · similarity 0.87)
- Distributive Fairness in Large Language Models: Evaluating Alignment with Human Values (NeurIPS 2025 · similarity 0.87)
- All Should Be Equal in the Eyes of LMs: Counterfactually Aware Fair Text Generation (AAAI 2024 · similarity 0.87)
- First-Person Fairness in Chatbots (ICLR 2025 · similarity 0.86)
- AIM-Fair: Advancing Algorithmic Fairness via Selectively Fine-Tuning Biased Models with Contextual Synthetic Data (CVPR 2025 · similarity 0.86)
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.
