The Partition Principle Revisited: Non-Equal Volume Designs Achieve Minimal Expected Star Discrepancy

This is a batch of recent ML/stat papers (arXiv 2026-03-03) covering advances across generative modeling (initialization-aware diffusion sampling), latent-space optimization (time-aware LSBO), uncertainty and reliability (Locus loss-quantile scoring; Deconditional Gaussian Processes; Co-optimization for Adaptive Conformal Prediction), evaluation and aggregation of LLM judges (CARE), causal inference for interacting human–AI systems with unobserved unit types, theoretical analyses (random features for operator-valued kernels; grokking via singular learning theory), and several applied methods (