Ziyuan Liu

Eliminating the need for per-scene optimization, this method boosts 3D model performance while enhancing visual fidelity and spatial consistency.

Frontier search agent performance does not require complex multi-agent swarms or test-time search verifiers: a single ReAct policy trained via iterative SFT-RL climbing hits 56.4% on Humanity's Last Exam and 92.9% on DeepSearchQA.

Uncertainty-calibrated MOPD boosts general capabilities in specialized language models by over 10% while maintaining domain-specific performance.

MoNO achieves unprecedented per-prompt diversity in diffusion sampling while eliminating the need for auxiliary quality-control objectives.

Change detection in remote sensing just got smarter: JL1-CC&QA not only identifies what changed but also explains why it matters through interactive questioning.

Surprisingly, the "think before answer" paradigm fails to enhance generative recommendation models, prompting a novel approach that redefines how reasoning is integrated into these systems.

Achieve superior control over the distortion-perception tradeoff in diffusion-based inverse problems by decoupling MAP estimation and posterior sampling into distinct stages.

Domain-specialized LLMs can regain lost general skills without sacrificing their expertise, thanks to a new distillation method that disentangles conflicting training signals.

Text-based speculative decoding falls flat for vision-language models, but ViSkip dynamically adapts to vision tokens for state-of-the-art acceleration.

By fusing confidence-weighted point cloud projections with a Kalman-inspired update mechanism, ConfCtrl enables diffusion models to generate geometrically consistent novel views from sparse inputs, even under significant viewpoint shifts.

MiroFlow leapfrogs existing LLM agent frameworks with its agent graph architecture, delivering state-of-the-art performance and robust execution across a diverse range of benchmarks.