Brief

Researchers have published a paper detailing advancements in McDiarmid's inequality, a tool applicable to statistics, learning theory, and theoretical computer science. The work highlights how approximate tensorization of entropy (ATE) implies McDiarmid's inequality and derives a version for non-isotropic Gaussian random vectors. The findings also extend concentration inequalities to strongly log-concave and log-smooth probability measures, improving upon prior results for non-i.i.d. observations. AI

A new research note, co-authored with AI assistance from Google's Gemini 3.5 Flash, presents a dimension-independent subgaussian concentration bound for Gaussian vectors under nonlinear mappings. This finding is applicable to any bounded function with a well-conditioned covariance. The researchers utilized this tool to address a specific question regarding sign-quantized linear maps. AI

Researchers have developed new methods for optimizing differentially private machine learning. One paper introduces a shuffling-aware optimization approach for private vector mean estimation, demonstrating that standard local differential privacy mechanisms can be suboptimal after shuffling. Another study proposes an optimal differentially private kernel learning algorithm using random projection, achieving minimax-optimal excess risk rates. Additionally, a third paper analyzes high-dimensional private linear regression, showing that practical algorithmic choices like gradient clipping and decaying learning rates lead to optimal risk rates. AI