Wei Liu

Achieve reliable segmentation of radiotherapy-induced normal tissue injuries, even with limited data, by intelligently prompting SAM with task-specific text, dose information, and iterative clicks.

Seedance 2.0 leapfrogs existing models by unifying multi-modal inputs (text, image, audio, video) into a single architecture for generating high-quality, longer-duration audio-video content.

By injecting clinically-relevant information like target contours and dose distributions, this registration framework achieves superior accuracy and robustness in longitudinal CT scans for proton therapy, outperforming generic deep learning methods.

Overcome the data scarcity that plagues antibody-antigen binding affinity prediction with a new method that turns the problem into a ranking task and improves top-candidate precision by 10%.

Existing deepfake detectors fall apart when faced with realistic, multi-step editing pipelines, highlighting the need for methods that can trace the *history* of manipulations, not just detect a single forgery.

Stop wasting human time on frontend code refinement: a VLM-powered critic can automatically guide LLMs to generate significantly better web UIs.

Federated learning can overcome data sparsity and privacy concerns to improve livestock growth prediction using real-world farm data.

Open LLMs can now rival proprietary systems in multilingual translation, as demonstrated by MiLMMT-46's competitive performance against Google Translate and Gemini 3 Pro.

Mobile GUI agents, despite benchmark successes, stumble badly when faced with the messy reality of third-party content, failing almost half the time.