Chandra Sekhar Seelamantula

Fine-tuning SOTA denoisers on texture datasets causes catastrophic forgetting, revealing a fundamental disconnect between general and texture-specific image restoration.

A closed-form, covariance-aware orthogonal transform for LLM quantization achieves state-of-the-art performance with a third of the memory and time of learned rotation methods.

This work derives the generalized score matching objective on a convex subset of $\mathbb{R}^{d}$ constructively starting from Minimum Probability Flow (MPF) learning, and shows how classical score matching as well as domain-adapted variants for non-negative data arise naturally within the proposed framework.

Dual-pixel sensors unlock surprisingly effective single-image fence removal, outperforming existing methods by fusing geometric and frequency-based priors.