SAHMM-VAE: A Source-Wise Adaptive Hidden Markov Prior Variational Autoencoder for Unsupervised Blind Source Separation

Collection of recent arXiv stat/ML papers with several items of security relevance. Notable entries: (1) Privacy-Accuracy Trade-offs in High-Dimensional LASSO analyzes differential-privacy mechanisms (output vs objective perturbation) and reveals non‑trivial failure modes that could weaken privacy guarantees if misapplied. (2) KANEL (virtual screening ensemble) can materially accelerate early hit enrichment in drug-discovery pipelines — potential biosecurity dual‑use risk if repurposed to prioritize harmful agents. (3) SAHMM‑VAE (source‑wise adaptive HMM priors) enables unsupervised blind‑-mix