Xian Li

UniPhysGen achieves state-of-the-art performance in grounding physical properties of 3D assets, enabling seamless integration into robotic simulations.

User-level depression detection can be dramatically improved by routing individuals to specialized experts based on weak semantic priors, rather than relying on a one-size-fits-all classifier.

Conditional mixup bridges the gap between pseudo-labeling and contrastive learning, setting a new benchmark in sound event detection performance.

Meta-optimizing an AI data scientist can dramatically enhance the quality of synthetic datasets, outperforming traditional methods.

PF-Trans achieves a remarkable PSNR of 48.50 dB, setting a new benchmark for spectral reconstruction in remote sensing.

Reflectance variability in UAV multispectral imagery can exceed 137% across viewing angles, challenging assumptions about radiometric consistency in remote sensing.

Illumination variations can be tackled without prior metadata, enabling precise anomaly detection in dense forest environments.

Attention mechanisms in a two-stream framework boost classification accuracy of airborne multispectral point clouds by effectively integrating spatial-spectral features.

Explicitly modeling occlusion intensity boosts UAV 3D object detection performance by over 9% in challenging environments.

On-policy reward modeling with LLM judges not only unlocks significant performance gains on complex mathematical reasoning tasks, but also generalizes to improve performance on simpler numerical and multiple-choice benchmarks.