Research map / Vision-language and multimodal / Multimodal LLMs
Personalizing MLLMs via Reinforced Multimodal Reference Game
Deepayan Das, Davide Talon, Yiming Wang, Massimiliano Mancini, Elisa Ricci
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 |
|---|---|---|---|
| train speaker with reference game (RRG) vs zero-shot speaker (ablation) | Qwen2-VL-7B · MyVLM captioning | F1: 89.3 -> 95.0 | Improved |
| train speaker with reference game vs zero-shot speaker (ablation) | Qwen2-VL-7B · Yo'LLaVA recognition | Wtd: 89.3 -> 94.7 | Improved |
| train speaker instead of listener (ablation) | RRG · MyVLM recognition | Wtd: 89.7 -> 91.8 | Improved |
| train speaker instead of listener (ablation) | RRG · Yo'LLaVA captioning | F1: 86.3 -> 88.6 | Improved |
| add reference game with distractors vs pairwise speaker-training (ablation) | RRG · MyVLM captioning | F1: 92.3 -> 95.0 | Improved |
| soft normalized reward instead of binary reward (ablation) | RRG · MyVLM captioning | F1: 93.4 -> 95.0 | Improved |
| soft normalized reward instead of binary reward (ablation) | RRG · Yo'LLaVA captioning | F1: 85.5 -> 88.6 | Improved |
| soft normalized reward instead of binary reward (ablation) | RRG · MyVLM recognition | Wtd: 92.3 -> 91.8 | Hurt |
| use hard instead of easy candidates (ablation) | RRG · MyVLM captioning | F1: 92.7 -> 95.0 | Improved |
| use hard instead of half-easy candidates (ablation) | RRG · MyVLM captioning | F1: 93.8 -> 95.0 | Improved |
| add hard negatives to hard positives (ablation) | RRG · MyVLM captioning | F1: 92.1 -> 95.0 | Improved |
| add hard positives to hard negatives (ablation) | RRG · MyVLM captioning | F1: 92.7 -> 95.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- RePIC — RePIC: Reinforced Post-Training for Personalizing Multi-Modal Language Models (NeurIPS 2025) · better on 3 of 4 shared results
- Yo'LLaVA — Yo'LLaVA: Your Personalized Language and Vision Assistant (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
- PersonaVLM: Long-Term Personalized Multimodal LLMs (CVPR 2026 · similarity 0.88)
- MM-R1: Unleashing the Power of Unified Multimodal Large Language Models for Personalized Image Generation (AAAI 2026 · similarity 0.87)
- RAP: Retrieval-Augmented Personalization for Multimodal Large Language Models (CVPR 2025 · similarity 0.86)
- Perception-R1: Advancing Multimodal Reasoning Capabilities of MLLMs via Visual Perception Reward (ICLR 2026 · similarity 0.86)
- RePIC: Reinforced Post-Training for Personalizing Multi-Modal Language Models (NeurIPS 2025 · similarity 0.86)
- MMPB: It’s Time for Multi-Modal Personalization (NeurIPS 2025 · similarity 0.85)
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.
