Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2410.16198 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913557938962432 |
|---|---|
| author | Zhang, Ruohong Zhang, Bowen Li, Yanghao Zhang, Haotian Sun, Zhiqing Gan, Zhe Yang, Yinfei Pang, Ruoming Yang, Yiming |
| author_facet | Zhang, Ruohong Zhang, Bowen Li, Yanghao Zhang, Haotian Sun, Zhiqing Gan, Zhe Yang, Yinfei Pang, Ruoming Yang, Yiming |
| contents | Chain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpretability and trustworthiness. However, current training recipes lack robust CoT reasoning data, relying on datasets dominated by short annotations with minimal rationales. In this work, we show that training VLM on short answers does not generalize well to reasoning tasks that require more detailed responses. To address this, we propose a two-fold approach. First, we distill rationales from GPT-4o model to enrich the training data and fine-tune VLMs, boosting their CoT performance. Second, we apply reinforcement learning to further calibrate reasoning quality. Specifically, we construct positive (correct) and negative (incorrect) pairs of model-generated reasoning chains, by comparing their predictions with annotated short answers. Using this pairwise data, we apply the Direct Preference Optimization algorithm to refine the model's reasoning abilities. Our experiments demonstrate significant improvements in CoT reasoning on benchmark datasets and better generalization to direct answer prediction as well. This work emphasizes the importance of incorporating detailed rationales in training and leveraging reinforcement learning to strengthen the reasoning capabilities of VLMs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_16198 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Improve Vision Language Model Chain-of-thought Reasoning Zhang, Ruohong Zhang, Bowen Li, Yanghao Zhang, Haotian Sun, Zhiqing Gan, Zhe Yang, Yinfei Pang, Ruoming Yang, Yiming Artificial Intelligence Computer Vision and Pattern Recognition 68T07 Chain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpretability and trustworthiness. However, current training recipes lack robust CoT reasoning data, relying on datasets dominated by short annotations with minimal rationales. In this work, we show that training VLM on short answers does not generalize well to reasoning tasks that require more detailed responses. To address this, we propose a two-fold approach. First, we distill rationales from GPT-4o model to enrich the training data and fine-tune VLMs, boosting their CoT performance. Second, we apply reinforcement learning to further calibrate reasoning quality. Specifically, we construct positive (correct) and negative (incorrect) pairs of model-generated reasoning chains, by comparing their predictions with annotated short answers. Using this pairwise data, we apply the Direct Preference Optimization algorithm to refine the model's reasoning abilities. Our experiments demonstrate significant improvements in CoT reasoning on benchmark datasets and better generalization to direct answer prediction as well. This work emphasizes the importance of incorporating detailed rationales in training and leveraging reinforcement learning to strengthen the reasoning capabilities of VLMs. |
| title | Improve Vision Language Model Chain-of-thought Reasoning |
| topic | Artificial Intelligence Computer Vision and Pattern Recognition 68T07 |
| url | https://arxiv.org/abs/2410.16198 |