Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Su, Zhaolong, Lu, Wang, Chen, Hao, Li, Sharon, Wang, Jindong
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2511.19413
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912987166539776
author Su, Zhaolong
Lu, Wang
Chen, Hao
Li, Sharon
Wang, Jindong
author_facet Su, Zhaolong
Lu, Wang
Chen, Hao
Li, Sharon
Wang, Jindong
contents Unified Multimodal Models (UMMs) have shown impressive performance in both understanding and generation with a single architecture. However, UMMs still exhibit a fundamental inconsistency: understanding favors compact embeddings, whereas generation favors reconstruction-rich representations. This structural trade-off produces misaligned decision boundaries, degraded cross-modal coherence, and heightened vulnerability under distributional and adversarial shifts. In this paper, we present UniGame, a self-adversarial post-training framework that directly targets the inconsistencies. By applying a lightweight perturber at the shared token interface, UniGame enables the generation branch to actively seek and challenge fragile understanding, turning the model itself into its own adversary. Experiments demonstrate that UniGame significantly improves the consistency (+4.6%). Moreover, it also achieves substantial improvements in understanding (+3.6%), generation (+0.02)on GenEval, out-of-distribution and adversarial robustness (+4.8% and +6.2% on NaturalBench and AdVQA). The framework is architecture-agnostic, introduces less than 1% additional parameters, and is complementary to existing post-training methods. These results position adversarial self-play as a general and effective principle for enhancing the coherence, stability, and unified competence of future multimodal foundation models. The official code is available at: https://github.com/AIFrontierLab/TorchUMM
format Preprint
id arxiv_https___arxiv_org_abs_2511_19413
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle UniGame: Turning a Unified Multimodal Model Into Its Own Adversary
Su, Zhaolong
Lu, Wang
Chen, Hao
Li, Sharon
Wang, Jindong
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Unified Multimodal Models (UMMs) have shown impressive performance in both understanding and generation with a single architecture. However, UMMs still exhibit a fundamental inconsistency: understanding favors compact embeddings, whereas generation favors reconstruction-rich representations. This structural trade-off produces misaligned decision boundaries, degraded cross-modal coherence, and heightened vulnerability under distributional and adversarial shifts. In this paper, we present UniGame, a self-adversarial post-training framework that directly targets the inconsistencies. By applying a lightweight perturber at the shared token interface, UniGame enables the generation branch to actively seek and challenge fragile understanding, turning the model itself into its own adversary. Experiments demonstrate that UniGame significantly improves the consistency (+4.6%). Moreover, it also achieves substantial improvements in understanding (+3.6%), generation (+0.02)on GenEval, out-of-distribution and adversarial robustness (+4.8% and +6.2% on NaturalBench and AdVQA). The framework is architecture-agnostic, introduces less than 1% additional parameters, and is complementary to existing post-training methods. These results position adversarial self-play as a general and effective principle for enhancing the coherence, stability, and unified competence of future multimodal foundation models. The official code is available at: https://github.com/AIFrontierLab/TorchUMM
title UniGame: Turning a Unified Multimodal Model Into Its Own Adversary
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2511.19413