Qianli Ma (original) (raw)

Qianli Ma I am a Ph.D. student at Shanghai Jiao Tong University, advised by Prof. Zhipeng Zhang . I completed my master's degree in Computer Science at Shanghai Jiao Tong University advised by Prof. Li Niu and Prof. Linfeng Zhang and my bachelor's degree in Instrument Science and Control Technology at Southeast University. I'm interested in generative AI such as diffusion models, LLMs. I am also interested in multimodal learning. I am actively seeking collaborations on exploring visual generation or reasoning ability of LLMs, please feel free to contact me!! Google Scholar / GitHub / Twitter / Email

"What I cannot create, I do not understand."

📑 Publications

* denotes equal contribution, † denotes corresponding author, some are highlighted.

	Decouple-Then-Merge: Finetune Diffusion Models as Multi-Task Learning Qianli Ma, Xuefei Ning, Dongrui Liu, Li Niu†, Linfeng Zhang† IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025 arXiv / PDF / BibTeX / Project Page / Code This paper proposes a new finetuning method for diffusion models, which decouples the diffusion process into multiple denoising tasks and then merges them. We show that this method can effectively finetune diffusion models for various tasks, including text-to-image generation, unconditional image generation.
	Efficient Diffusion as Low Light Enhancer Guanzhou Lan, Qianli Ma, Yuqi Yang, Zhigang Wang, Dong Wang, Xuelong Li†, Bin Zhao† IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025 arXiv / PDF / BibTeX / Project Page / Code This paper proposes an efficient diffusion model for low light enhancement, which can be applied to various low light enhancement tasks.
	LED-Merging: Mitigating Safety-Utility Conflicts in Model Merging with Location-Election-Disjoint Qianli Ma, Dongrui Liu, Qian Chen, Linfeng Zhang, Jing Shao† The 63rd Annual Meeting of the Association for Computational Linguistics (ACL main), 2025 arXiv / BibTeX / Code This paper proposes a method to mitigate safety-utility conflicts in model merging for LLMs, which can be applied to various safety-utility tasks.
	VBench++: Comprehensive and Versatile Benchmark Suite for Video Generative Models Ziqi Huang, Fan Zhang, Xiaojie Xu, Yinan He, Jiashuo Yu, Ziyue Dong, Qianli Ma, Nattapol Chanpaisit, Chenyang Si, Yuming Jiang, Yaohui Wang, Xinyuan Chen, Ying-Cong Chen, Limin Wang, Dahua Lin†, Yu Qiao†, Ziwei Liu† ArXiv, 2024 arXiv / BibTeX / Project Page / Code This paper proposes a comprehensive and versatile benchmark suite for video generative models.
	Token Pruning for Caching Better: 9 Times Acceleration on Stable Diffusion for Free Evelyn Zhang, Bang Xiao, Fufu Yu, Jiayi Tang, Chang Zou, Ke Yan, Shouhong Ding, Qianli Ma, Fei Ren, Linfeng Zhang† ArXiv, 2025 arXiv / BibTeX / Code This paper proposes a token pruning method for stable diffusion, which can accelerate the generation process by 9 times.