Discrete Flow Maps

Collection of April 14, 2026 machine-learning papers covering advances in generative modelling, robust/statistical estimation, causal/precision-health methods, and interpretability. Key contributions include Discrete Flow Maps for single-step discrete sequence generation aligned with simplex geometry; orthogonal/DR and R-learners generalized to conditional odds and risk ratios; deep generative frameworks for stratified data (mixtures of VAEs and diffusion-based score methods); One-step Score-based Density Ratio Estimation (OS-DRE) for solver-free DRE; tail-aware information-theoretic generaliz