Recent Developments in Data-Driven Algorithms for Discrete Optimization

The last few years have witnessed a renewed interest in “data-driven algorithm design” (Balcan 2020), the use of Machine Learning (ML) to tailor an distribution instances. More than decade ago, advances configuration (Hoos 2011) paved way for historical data modify algorithm’s (typically fixed, static) parameters. In discrete optimization (e.g., satisfiability, integer programming, etc.), exact and inexact algorithms NP-Hard problems often involve heuristic search decisions (Lodi 2013), abstracted as parameters, that can demonstrably benefit from tuning on instances application interest. While useful, may be insufficient: setting parameters upfront solving input instance is still static, high-level decision. contrast, we been exploring suite ML Reinforcement (RL) approaches tune iterative algorithms, such branch-and-bound programming or construction heuristics, at iteration-level (Khalil et al. 2016, 2017; Dai Chmiela 2021; Gupta 2022; Chi Khalil, Vaezipoor, Dilkina Morris, Lodi Alomrani, Moravej, Khalil Cappart 2020). We will survey our most recent work this area: 1. New methods learning MILP (Gupta 2020, 2022); 2. RL online combinatorial largescale linear (Alomrani, 3. Neural network approximations stochastic (Dumouchelle 2022).