Kaushik Roy

REALIGN: Regularized Procedure Alignment with Matching Video Embeddings via Partial Gromov-Wasserstein Optimal Transport

KVLinC: KV Cache Quantization with Hadamard Rotation and Linear Correction

Memorization Through the Lens of Sample Gradients

LANCE: Low Rank Activation Compression for Efficient On-Device Continual Learning

TopoCore: Unifying Topology Manifolds and Persistent Homology for Data Pruning

SlimDiff: Training-Free, Activation-Guided Hands-free Slimming of Diffusion Models

CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Models

Towards Memorization Estimation: Fast, Formal and Free

DCT-CryptoNets: Scaling Private Inference in the Frequency Domain

ResQ: Mixed-Precision Quantization of Large Language Models with Low-Rank Residuals

Unlocking SVD-Space for Feedback Aligned Local Training

Memorization and the Orders of Loss: A Learning Dynamics Perspective

The Relevancy Metric: Understanding the Impact of Training Data