Yu Liu

SemGeoNav achieves superior obstacle avoidance and navigation efficiency by merging semantic reasoning with geometric planning, outperforming leading methods in real-world tests.

Ling-2.6 and Ring-2.6 achieve unprecedented efficiency in agentic intelligence, enabling instant responses and deep reasoning at trillion-parameter scale.

Language complexity is not just about length; the ladderpath index reveals deep structural invariances across diverse languages.

Code generation can be both private and high-utility, with PrivCode++ achieving superior results by safeguarding both prompts and snippets.

Nova Teacher achieves a remarkable 5.22% improvement in source detection accuracy, setting a new standard for semi-supervised learning in astronomy.

Generating realistic 3D environments from satellite imagery in under 10 minutes could revolutionize how we visualize and interact with our planet.

Open-source E2E models outperform Cascade systems in understanding heavy dialects, but struggle significantly with low-resource languages, revealing critical gaps in current speech model architectures.

LLMs can drive significant gains in ad systems, not just as rankers or retrievers, but as complementary predictors that forecast likely advertisers from user profiles.

Long video generation fails not just because of limited context length, but because of *how* that context is allocated – and ReCA's hierarchical approach shows a way to fix it.

Fine-tuning LLMs in federated settings just got easier: SplitFT lets clients adapt their cut layers and LoRA ranks, boosting performance and slashing communication costs.

Robots can now navigate complex outdoor environments using only high-level human instructions and readily available GPS/map data, bypassing the need for expensive HD maps or limited short-horizon policies.

Skip the bulky bidirectional teacher: this new method trains a fast, causal audio-video generator directly, slashing sampling steps while maintaining top-tier quality.