Flow Matching Framework on Riemannian Symmetric Spaces
Discover a novel flow matching framework leveraging Riemannian symmetric spaces for efficient machine learning and complex data transformations.
PatRe: Benchmark for Patent Office Actions & Rebuttals
Discover PatRe, the first full-stage benchmark modeling patent office actions and rebuttals to improve patent examination efficiency.
Orthogonal Task Decomposition for Multi-Modal Clinical Data
Discover how Orthogonal Task Decomposition improves multi-task learning from multi-modal clinical data, boosting prediction accuracy in healthcare.
HeadQ: Optimizing KV-Cache Quantization for AI Models
Discover how HeadQ improves KV-cache quantization using model-visible distortion and score-space correction for enhanced AI model efficiency.
PerFlow: Efficient Physics-Based Reconstruction of Spatiotemporal Dynamics
Discover PerFlow, a novel physics-embedded method for fast, accurate reconstruction and uncertainty quantification of spatiotemporal PDE systems from spars...