Research map / Multi-agent systems and games / Games and equilibria
Tackling Model Bias via Game-theoretic Multi-agent Collaboration Framework for Hateful Meme Classification
Yiwei Wei, Zhengliang Guo, Shaozu Yuan, Chengyin Hu, Zhiyang Jia, Jiujiang Guo, Meng Chen, Peiying Wang, Longbiao Wang
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 |
|---|---|---|---|
| keep master (classification) agent in the game (ablation) | GECO · PrideMM | accuracy: 62.88 -> 82.84 | Improved |
| add LLaVA reasoning agent (ablation) | GECO · PrideMM | accuracy: 81.66 -> 82.84 | Improved |
| add CLIP learnable agent (ablation) | GECO · MultiOff | accuracy: 75.17 -> 78.52 | Improved |
| add Qwen reasoning agent (ablation) | GECO · MultiOff | accuracy: 73.83 -> 78.52 | Improved |
| add Gemma reasoning agent (ablation) | GECO · MultiOff | accuracy: 75.84 -> 78.52 | Improved |
| keep both agents of a pair rather than removing both (ablation) | GECO · PrideMM | accuracy: 79.88 -> 82.84 | Improved |
| add pairwise and all-hit cooperative bonuses to individual bonus (ablation) | GECO · PrideMM and MultiOff | accuracy | Improved |
| optimize LMM with game-theoretic GECO instead of supervised fine-tuning | Gemma3-4B · MultiOff | accuracy: 65.10 -> 76.51 | Improved |
| optimize LMM with game-theoretic GECO instead of SFT | LLaVA-1.5-13B · MultiOff | F1: 58.62 -> 57.10 | Hurt |
| optimize LMM with game-theoretic GECO instead of SFT | LLaVA-1.5-13B · PrideMM | accuracy: 78.70 -> 81.45 | Improved |
| replace voting ensemble with GECO | multi-agent ensemble · MultiOff | accuracy: 69.80 -> 78.52 | Improved |
| replace debate ensemble with GECO | multi-agent ensemble · PrideMM | accuracy: 79.88 -> 82.84 | Improved |
Closest papers by meaning
- TRACE: Textual Relevance Augmentation and Contextual Encoding for Multimodal Hate Detection (AAAI 2026 · similarity 0.86)
- Hateful Meme Detection through Context-Sensitive Prompting and Fine-Grained Labeling (Student Abstract) (AAAI 2025 · similarity 0.86)
- Private Blotto: Viewpoint Competition with Polarized Agents (AAAI 2025 · similarity 0.84)
- LLM Active Alignment: A Nash Equilibrium Perspective (NeurIPS 2026 · similarity 0.84)
- Incentivizing Truthful Language Models via Peer Elicitation Games (NeurIPS 2025 · similarity 0.84)
- DEBATE: A Large-Scale Benchmark for Evaluating Opinion Dynamics in Role-Playing LLM Agents (NeurIPS 2026 · similarity 0.84)
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.
