Huan Yu

R^3 achieves a groundbreaking balance between compliance and semantic intent preservation, outperforming existing methods in video ad rectification.

Explainable detection of hateful videos is now possible, revealing the nuanced reasoning behind classifications that traditional methods overlook.

Stop sacrificing subject fidelity for editability: DisCo lets you have both in text-to-image generation by disentangling and recoupling visual and textual information.

Current subject-driven text-to-image models struggle with specific subject categories and prompt scenarios, a problem exposed by a new benchmark that also offers actionable insights for improvement.

OneRanker tackles the optimization tension in generative recommendation systems by deeply integrating generation and ranking into a single architecture, achieving state-of-the-art performance in industrial advertising.

Dramatically reduce hallucination in industrial RAG systems by jointly optimizing retrieval and generation with graph-aware retrieval and reinforcement learning, leading to a 92.7% reduction in URL hallucination in a real-world advertising QA system.

Agentic RAG gets a 7.7 point accuracy boost thanks to Search-P1's path-centric reward shaping, which extracts learning signals even from failed reasoning attempts.

LLMs still struggle with real-world advertising analytics, with even Gemini-3-Pro dropping to 49.4% accuracy on the most complex tasks in the new AD-Bench benchmark.

Spatial relationship hallucinations in image inpainting can be significantly reduced by directly optimizing for preferences on background plausibility, even when foregrounds are identical.