LoBoost: Fast Model-Native Local Conformal Prediction for Gradient-Boosted Trees

This is a collection of recent ML/statistics preprints (Feb 27, 2026) covering topics with operational security and privacy relevance: model-native conformal prediction for uncertainty (LoBoost); theoretical/algorithmic advances in generative modeling (flow matching, masked diffusion, sampling error bounds); Bayesian neural network / GP scaling and posterior behavior; continual unsupervised fine-tuning for amortized Bayesian inference (addresses catastrophic forgetting); differential-privacy improvements via public-moment-guided truncation (PMT); decentralized consensus via gossip (Borda/Cop e