Xin Zhang

Reinforcement learning can significantly enhance OOD detection by reducing false positives in evolving environments, outperforming traditional gradient descent methods.

GEAR-VLA achieves a remarkable 90.1% success rate in universal grasping tasks, showcasing its ability to generalize across unseen objects and diverse robot embodiments.

NKW redefines narrative QA by seamlessly integrating story dynamics and character interactions, achieving superior performance in understanding complex narratives.

PHKT outperforms traditional models by effectively capturing user-specific behaviors and preferences through a novel hypergraph approach and advanced temporal modeling.

Forget data selection—reordering your existing dataset using these four simple guidelines can significantly boost LLM training performance and stability.

LLM agents learn tool use far more efficiently when their learning environment evolves alongside them, adapting to the agent's changing capabilities.

Forget hours-long simulations: EnergAIzer slashes GPU power estimation time to seconds while maintaining accuracy, by exploiting structured patterns in AI kernel optimizations.

Real-world proactive agents can now infer latent user needs and act on them in real-time, rivaling state-of-the-art models in intent detection while maintaining low latency.

Knowing the *perfect* API to use or *exact* location to edit could drastically improve SWE agent performance, but knowing the perfect regression test result? Not so much.

LLM agents become *less* like humans as you crank up the reasoning, unless you explicitly model value activation with a Value Verifier trained on real-world interaction data.

By combining CNNs and State Space Models, DA-Mamba achieves efficient global-local feature alignment for domain adaptive object detection, outperforming prior CNN-only and Transformer-based approaches.