Zi-Xuan Wang

Robotic recovery is now 74-93% faster by triggering corrections from action proposal disagreement and using efficient visual residuals, not slow VLM reasoning.

Explicit 3D geometry, rendered as text, unlocks superior spatial reasoning in RGB-only VLMs by guiding trajectory correction during distillation.

Scaling quantum chemistry across modular QPUs hinges on ansatz selection: pairing UpCCD with spin-blocked Jordan-Wigner encoding dramatically slashes inter-chip Bell-pair consumption compared to the prohibitive overhead of standard UCCSD.

Modifying executable agent harnesses rather than model weights allows test-time adaptation to generalize across unseen reasoning tasks without a single gradient step.

Gaussian guidance reveals that optimal trajectory retention depth varies across tasks, leading to a substantial performance boost in reinforcement learning.

Looping discrete embeddings with continuous hidden states enables near-perfect accuracy in multi-hop reasoning with fewer training steps than traditional methods.

A novel framework achieves unprecedented dataset distillation speed and accuracy by directly minimizing information loss, setting a new benchmark in the field.

Decoupling perception from reasoning in visual tasks leads to a remarkable 93.2% accuracy on V-Star, showcasing a new paradigm for fine-grained visual reasoning.

Memory recovery in LLM agents is not just a byproduct of task success; it's a distinct capability that remains underexplored, with current models showing only moderate performance in reconstructing user states.

A novel multi-agent framework enhances zero-shot 3D understanding by iteratively optimizing viewpoints and integrating fragmented observations, leading to significant performance gains.

Action chunk utilization triples and physical execution steps drop by over 50%, resulting in a 5.83x speedup in VLA model deployment without sacrificing performance.

LLMs can escape the trap of confidently wrong reasoning by co-evolving a generator and verifier from a single model, bootstrapping each other to break free from flawed consensus.