Attention Reallocation: Towards Zero-cost and Controllable Hallucination Mitigation of MLLMs
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866929755251539968 |
|---|---|
| author | Tu, Chongjun Ye, Peng Zhou, Dongzhan Bai, Lei Yu, Gang Chen, Tao Ouyang, Wanli |
| author_facet | Tu, Chongjun Ye, Peng Zhou, Dongzhan Bai, Lei Yu, Gang Chen, Tao Ouyang, Wanli |
| contents | Multi-Modal Large Language Models (MLLMs) stand out in various tasks but still struggle with hallucinations. While recent training-free mitigation methods mostly introduce additional inference overhead via retrospection strategy and contrastive decoding, we propose attention reallocation (AttnReal) to mitigate hallucinations with nearly zero extra cost. Our approach is motivated by the key observations that, MLLM's unreasonable attention distribution causes features to be dominated by historical output tokens, which further contributes to hallucinated responses because of the distribution gap between different token types. Based on the observations, AttnReal recycles excessive attention from output tokens and reallocates it to visual tokens, which reduces MLLM's reliance on language priors and ensures the decoding process depends more on the visual inputs. More interestingly, we find that, by controlling the intensity of AttnReal, we can achieve a wide-range trade-off between the response faithfulness and overall performance. Comprehensive results from different benchmarks validate the effectiveness of AttnReal across six open-source MLLMs and three decoding strategies. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_08342 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Attention Reallocation: Towards Zero-cost and Controllable Hallucination Mitigation of MLLMs Tu, Chongjun Ye, Peng Zhou, Dongzhan Bai, Lei Yu, Gang Chen, Tao Ouyang, Wanli Computer Vision and Pattern Recognition Multi-Modal Large Language Models (MLLMs) stand out in various tasks but still struggle with hallucinations. While recent training-free mitigation methods mostly introduce additional inference overhead via retrospection strategy and contrastive decoding, we propose attention reallocation (AttnReal) to mitigate hallucinations with nearly zero extra cost. Our approach is motivated by the key observations that, MLLM's unreasonable attention distribution causes features to be dominated by historical output tokens, which further contributes to hallucinated responses because of the distribution gap between different token types. Based on the observations, AttnReal recycles excessive attention from output tokens and reallocates it to visual tokens, which reduces MLLM's reliance on language priors and ensures the decoding process depends more on the visual inputs. More interestingly, we find that, by controlling the intensity of AttnReal, we can achieve a wide-range trade-off between the response faithfulness and overall performance. Comprehensive results from different benchmarks validate the effectiveness of AttnReal across six open-source MLLMs and three decoding strategies. |
| title | Attention Reallocation: Towards Zero-cost and Controllable Hallucination Mitigation of MLLMs |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2503.08342 |