• Design CoT Supervision From Domain Experts' Actual Reasoning Process. In medical VQA, structured clinical workflows as CoT steps improve both accuracy and traceability. The approach transfers to any vertical requiring structured professional judgment. CVPR accepted.

• How Few Features Reproduce a Model's Rejection Decision? Framing minimum abductive explanation as 0-1 integer programming yields solutions faster than methods that don't guarantee minimality. Limited to linear models, but the problem framing matters for high-stakes human-AI collaboration.

• Train on Synthetic Novel Classes to Discover Real Ones at Inference. Ditching hash encoding for a pure feature-space approach eliminates the train-inference objective mismatch. Up to +13.1% all-class accuracy across seven benchmarks. CVPR accepted.

Also Notable

• NVIDIA Releases Long-Video Multimodal Understanding Benchmark MMOU — jointly tests visual, audio, and text reasoning, exposing current models' cross-modal joint reasoning gaps.
• Graph-Based Decomposition for Dynamic Routing in Time-Series Forecasting — balances Channel-Independent generalization with Channel-Dependent expressiveness. ICLR accepted.
• Distinguishing Error Severity in Pathology Slide Multi-Classification — misclassifying benign as malignant and missing malignancy carry very different costs. CVPR accepted.
• Polarization Cues Constrain Physical Properties in 3D Gaussian Reconstruction — improves albedo and normal estimation for reflective objects.
• Efficient RGB 3D Reconstruction Pipeline for Visually Repetitive Environments — targets automated infrastructure inspection. ICCV accepted.

Read the full edition →