Decision-Focused Learning: Foundations, State of the Art, Benchmark and Future Opportunities
Decision-Focused Learning: Foundations, State of the Art, Benchmark and Future Opportunities
复制标题
DOI：
10.48550/arxiv.2307.13565
复制
发表时间：
2023-07
期刊：
影响因子：
--
通讯作者：
Jayanta Mandi;James Kotary;Senne Berden;Maxime Mulamba;Víctor Bucarey;Tias Guns;Ferdinando Fioretto
中科院分区：
文献类型：
--
作者：
Jayanta Mandi;James Kotary;Senne Berden;Maxime Mulamba;Víctor Bucarey;Tias Guns;Ferdinando Fioretto
以决策为中心的学习（DFL）是一种新兴的范式，它集成了机器学习（ML）和约束优化，通过在端到端系统中训练ML模型来提高决策质量。这种方法显示出巨大的潜力，革命性的组合决策在现实世界中的应用程序，在不确定性下运行，估计决策模型中的未知参数是一个重大挑战。本文对DFL进行了全面的回顾，深入分析了用于联合收割机ML和约束优化的基于梯度和无梯度技术。它评估了这些技术的优势和局限性，并包括对七个问题的十一种方法进行了广泛的实证评估。该调查还提供了DFL的最新进展和未来研究方向的见解。代码和基准：https://github.com/PredOpt/predopt-benchmarks
Decision-focused learning (DFL) is an emerging paradigm that integrates machine learning (ML) and constrained optimization to enhance decision quality by training ML models in an end-to-end system. This approach shows significant potential to revolutionize combinatorial decision-making in real-world applications that operate under uncertainty, where estimating unknown parameters within decision models is a major challenge. This paper presents a comprehensive review of DFL, providing an in-depth analysis of both gradient-based and gradient-free techniques used to combine ML and constrained optimization. It evaluates the strengths and limitations of these techniques and includes an extensive empirical evaluation of eleven methods across seven problems. The survey also offers insights into recent advancements and future research directions in DFL. Code and benchmark: https://github.com/PredOpt/predopt-benchmarks