Hao Sun

Early detection of failure in LLM agents can save over 47% of inference compute by leveraging internal representations rather than observable behavior.

Major LLM failures like hallucination and bias can be traced to specific disruptions in cognitive organization, offering a roadmap for targeted interventions.

MIRTH transforms VLA models by enabling them to effectively leverage historical dynamics and optimize action trajectories, leading to unprecedented error recovery in robotic control.

TryOnCrafter revolutionizes video virtual try-on by enabling dynamic, omnidirectional viewpoint exploration without sacrificing realism or structural integrity.

Kops enables significant performance improvements in eBPF by allowing new operations to be added without compromising kernel safety or increasing the trusted computing base.

GraphPO slashes redundancy in reasoning model training, enabling more efficient exploration and improved performance on complex tasks.

Aligning embodiment structures within a shared grasp generator boosts performance and transferability across diverse end effectors, challenging conventional design approaches.

MemVenom reveals that web agents can be compromised with up to 99.15% success through sophisticated memory poisoning attacks that bypass traditional defenses.

The new REO framework reveals that the true challenge in differential equation discovery lies not just in recovering equations, but in leveraging them to reshape scientific understanding.

Transforming any 2D representation into a symmetric form without losing continuity could revolutionize design processes across various fields.

Solving Poisson equations just got faster and more stable: NPSolver trains neural operators without solution labels by iteratively refining predictions with preconditioned conjugate gradient steps.

An AI agent autonomously discovered four new superconductors, shrinking the discovery timeline from years to GPU hours.