collaborators

8 papers

cs.LG2026

Adapting in the Dark: Efficient and Stable Test-Time Adaptation for Black-Box Models

Yunbei Zhang, Shuaicheng Niu, Chengyi Cai +2

Test-Time Adaptation (TTA) for black-box models accessible only via APIs remains a largely unexplored challenge. Existing approaches such as post-hoc output refinement offer limite…

cs.CV2026

Prime Once, then Reprogram Locally: An Efficient Alternative to Black-Box Service Model Adaptation

Yunbei Zhang, Chengyi Cai, Feng Liu +1

Adapting closed-box service models (i.e., APIs) for target tasks typically relies on reprogramming via Zeroth-Order Optimization (ZOO). However, this standard strategy is known for…

cs.LG2026

Per-parameter Task Arithmetic for Unlearning in Large Language Models

Chengyi Cai, Zesheng Ye, Jiangchao Yao +5

In large language model (LLM) unlearning, private information is required to be removed. Task arithmetic unlearns by subtracting a specific task vector (TV)--defined as the paramet…

cs.LG2026

Visual-Guided Key-Token Regularization for Multimodal Large Language Model Unlearning

Chengyi Cai, Zesheng Ye, Peike Li +3

Unlearning in Multimodal Large Language Models (MLLMs) prevents the model from revealing private information when queried about target images. Existing MLLM unlearning methods larg…

cs.CV2026

Let's Roll a BiFTA: Bi-refinement for Fine-grained Text-visual Alignment in Vision-Language Models

Yuhao Sun, Chengyi Cai, Jiacheng Zhang +3

Recent research has shown that aligning fine-grained text descriptions with localized image patches can significantly improve the zero-shot performance of pre-trained vision-langua…

cs.LG2025

Neural Network Reprogrammability: A Unified Theme on Model Reprogramming, Prompt Tuning, and Prompt Instruction

Zesheng Ye, Chengyi Cai, Ruijiang Dong +4

As large-scale pre-trained foundation models continue to expand in size and capability, efficiently adapting them to specific downstream tasks has become increasingly critical. Des…