Guang Yang

MLLMs can ace circuit-to-code generation by cheating with identifier semantics, so anonymizing those identifiers reveals a shocking lack of true visual grounding.

GaLa's hypergraph representation reveals hidden semantic relationships in multimodal data, leading to a dramatic boost in procedural planning accuracy.

CT foundation models perform better as feature extractors for lightweight probes than as vision encoders for vision-language models, even on tasks like report generation.

Forget treating document graphics as mere pixels: this new OCR system parses them into reusable code, unlocking multimodal supervision and outperforming existing systems.

Achieve ~20% gains on difficult 3D medical image segmentation by explicitly removing noisy activations in U-Net skip connections with a novel proximal-sparse attention mechanism.

Low-dose PET scans get a boost with MAP-Diff, a diffusion model that uses intermediate-dose scans as anchors, leading to sharper, more accurate reconstructions.