Salvato in:
| Autori principali: | , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2603.15724 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917348603068416 |
|---|---|
| author | Tan, Lit Sin Chen, Junzhe Fu, Xiaolong Ma, Lichen Huang, Junshi Shi, Jianzhong Li, Yan Wen, Lijie |
| author_facet | Tan, Lit Sin Chen, Junzhe Fu, Xiaolong Ma, Lichen Huang, Junshi Shi, Jianzhong Li, Yan Wen, Lijie |
| contents | Existing test-time scaling (TTS) methods for unified multimodal models (UMMs) in text-to-image (T2I) generation primarily rely on search or sampling strategies that produce only instance-level improvements, limiting the ability to learn from prior inferences and accumulate knowledge across similar prompts. To overcome these limitations, we propose Meta-TTRL, a metacognitive test-time reinforcement learning framework. Meta-TTRL performs test-time parameter optimization guided by model-intrinsic monitoring signals derived from the meta-knowledge of UMMs, achieving self-improvement and capability-level improvement at test time. Extensive experiments demonstrate that Meta-TTRL generalizes well across three representative UMMs, including Janus-Pro-7B, BAGEL, and Qwen-Image, achieving significant gains on compositional reasoning tasks and multiple T2I benchmarks with limited data. We provide the first comprehensive analysis to investigate the potential of test-time reinforcement learning (TTRL) for T2I generation in UMMs. Our analysis further reveals a key insight underlying effective TTRL: metacognitive synergy, where monitoring signals align with the model's optimization regime to enable self-improvement. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_15724 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Meta-TTRL: A Metacognitive Framework for Self-Improving Test-Time Reinforcement Learning in Unified Multimodal Models Tan, Lit Sin Chen, Junzhe Fu, Xiaolong Ma, Lichen Huang, Junshi Shi, Jianzhong Li, Yan Wen, Lijie Machine Learning Artificial Intelligence Existing test-time scaling (TTS) methods for unified multimodal models (UMMs) in text-to-image (T2I) generation primarily rely on search or sampling strategies that produce only instance-level improvements, limiting the ability to learn from prior inferences and accumulate knowledge across similar prompts. To overcome these limitations, we propose Meta-TTRL, a metacognitive test-time reinforcement learning framework. Meta-TTRL performs test-time parameter optimization guided by model-intrinsic monitoring signals derived from the meta-knowledge of UMMs, achieving self-improvement and capability-level improvement at test time. Extensive experiments demonstrate that Meta-TTRL generalizes well across three representative UMMs, including Janus-Pro-7B, BAGEL, and Qwen-Image, achieving significant gains on compositional reasoning tasks and multiple T2I benchmarks with limited data. We provide the first comprehensive analysis to investigate the potential of test-time reinforcement learning (TTRL) for T2I generation in UMMs. Our analysis further reveals a key insight underlying effective TTRL: metacognitive synergy, where monitoring signals align with the model's optimization regime to enable self-improvement. |
| title | Meta-TTRL: A Metacognitive Framework for Self-Improving Test-Time Reinforcement Learning in Unified Multimodal Models |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2603.15724 |