Manifold Constrained Conformal Prediction for Spatial Events

Collection of new arXiv stat-ML papers (14 Jul 2026) covering advances in uncertainty quantification, spatio-temporal forecasting, causal discovery, representation learning for mediation, short-text clustering, neural network approximation theory, robust sparse demixing, equivariant interatomic potentials, long-memory reservoir computing for epidemiological forecasting, and contextual bandits. Highlights: a manifold-constrained conformal prediction method for spatial point-cloud events using (sliced) Wasserstein scores and flow-based ensemble sampling; TSCoNet (CNN-LSTM + Gaussian copula) for: