Convolutional Surrogate for 3D Discrete Fracture-Matrix Tensor Upscaling

This document is an arXiv feed (multiple ML/CS papers) covering: a 3D convolutional surrogate for discrete fracture–matrix (DFM) hydraulic upscaling (geoscience/simulation); an autoregressive generative model that produces counterfactual patient timelines from real-world EHR data (300k patients, 400M events); LiME, a lightweight Mixture-of-Experts PEFT method for efficient multimodal multi-task learning; SIEVE, a sample-efficient method for parametric learning from natural language context; model-scheduling to speed masked diffusion language model sampling; PROGRS, a method to incorporate (and