One-Shot Generative Flows: Existence and Obstructions

Batch of arXiv stat/ml papers (announced 2026-04-20) covering theoretical and algorithmic advances across generative modeling, optimization, statistical learning, and applications. Key contributions include: analysis of one-shot generative flows and a dichotomy for straight-line flow existence under independent endpoints (existence for Gaussian targets, impossibility for well-separated multimodal targets); a PRIM-based PCA result establishing an unsupervised No-Free-Lunch phenomenon and algorithms for bump-hunting; provably optimal multi-fidelity optimization (Kometo) with improved rates; a軽)轻