Recently, Antoniadis et al. (ICLR 2025) proposed a framework for incorporating predictions to approximate NP-hard selection problems.
Learning-augmented algorithms combine fallible predictions with formal performance guarantees. This survey reviews prediction interfaces, error measures, consistency–robustness trade-offs, and five construction mechanisms across online optimization, caching, learned data structures, graph problems, and mechanism design. It distinguishes theorem-level upper bounds from matched asymptotic dependence, separates formal guarantees from empirical evidence, and outlines open problems in cost-aware prediction, endogenous error, semantic predictors, and benchmarking.
By Hailiang Zhao, Peng Chen, Xueyan Tang, Jianwei Yin, Shuiguang Deng
The paper investigates how algorithms for hard combinatorial optimization problems converge to trivial solutions, focusing on finite-size behavior rather than asymptotic limits. By analyzing large-graph asymptotics and running numerical experiments on problems like maximum independent set and maximum K‑SAT, the authors show that convergence to theoretically predicted bounds is surprisingly slow. In the intermediate regime of high constraint density, local algorithms actually outperform their asymptotic predictions, highlighting a gap between finite-regime performance and asymptotic theory.
arXiv:2606. 19587v1 Announce Type: cross Abstract: We propose a scalable method for training prediction (machine learning) models in the predict-then-optimize paradigm, where model outputs serve as coefficients for a subsequent linear optimization task.
By Beichen Wan, Mo Liu
arXiv:2606. 01342v1 Announce Type: cross Abstract: Learning-augmented paging has been extensively studied in recent years.
By Peng Chen, Hailiang Zhao, Xueyan Tang, Yixuan Wang, Shuiguang Deng
arXiv:2602. 03972v3 Announce Type: replace-cross Abstract: The best-arm identification (BAI) problem is one of the most fundamental problems in interactive machine learning, which has two flavors: the fixed-budget setting (FB) and the fixed-confidence setting (FC).
By Kapilan Balagopalan, Yinan Li, Yao Zhao, Tuan Nguyen, Anton Daitche, Houssam Nassif, Kwang-Sung Jun