Price of Fairness in Bandits: A Tight Minimax Characterization

Batch of arXiv July 16, 2026 ML/stat papers covering algorithmic and theoretical advances plus privacy/operational decision frameworks. Key items: (1) “Price of Fairness in Bandits” characterizes the exact polynomial cost of strict fairness for p<0 (q>0) bandit objectives and introduces UCB‑HARE matching lower bounds up to logs; (2) “Non‑Expansive Two‑Time‑Scale Stochastic Approximation” gives sharp lower bounds for Krasnoselskii–Mann residuals and proposes bias‑corrected / single‑loop accelerations improving rates; (3) “Parallel gradient boosting” proposes a common‑descent direction to scale/