Central Description Length (CDL) Clustering Validation Index

This arXiv feed (multiple new papers) presents a set of machine-learning contributions across unsupervised learning, imputation, optimization, transfer, and robust decision-making. Key highlights: (1) Central Description Length (CDL) — a new clustering validation index that better selects cluster counts on non‑convex and variable‑density data without kernel preprocessing; (2) HyFAD — a hybrid time/frequency diffusion model with frequency-aware embeddings achieving state‑of‑the‑art time‑series imputation (code released); (3) Deterministic envelopes for tamed SGLD — a taming design that avoids a