Tingyu Song

Cross-attentive MLLMs already solve the attribute-binding failures that cripple embedding models—and distilling their soft multi-level ranking distributions via Rank-KL unlocks that reasoning for bi-encoders without degrading standard recall.

Even the strongest Android GUI agents falter under dynamic runtime anomalies, revealing critical vulnerabilities in their design.

Despite advances in visual realism, models struggle with accurately capturing scientific reasoning and causal dynamics, revealing a critical gap in video generation capabilities.

UEmbed achieves a groundbreaking unification of dense and sparse embeddings, outperforming existing models in multimodal tasks while streamlining the retrieval process.

Skill augmentation can elevate CLI agent performance beyond GUI counterparts, revealing that the true challenge lies in skill coverage rather than model capability.

Models trained on the new VideoKR dataset achieve superior performance in knowledge-intensive video reasoning, setting a new benchmark for the field.

Standard retriever evaluations hide critical weaknesses in agentic search systems, but a new benchmark and training method exposes and addresses these flaws.