Accelerating Cutting-Plane Algorithms via Reinforcement Learning Surrogates

Authors: Kyle Mana, Fernando Acero, Stephen Mak, Parisa Zehtabi, Michael Cashmore, Daniele Magazzeni, Manuela Veloso

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our method on two types of problems where cutting-plane algorithms are commonly used: stochastic optimization, and mixed-integer quadratic programming. We observe the benefits of our method when applied to Benders decomposition (stochastic optimization) and iterative loss approximation (quadratic programming), achieving up to 45% faster average convergence when compared to modern alternative algorithms.
Researcher Affiliation Collaboration 1J.P. Morgan AI Research 2University College London 3University of Cambridge {kyle.mana, parisa.zehtabi, michael.cashmore, daniele.magazzeni, manuela.veloso}@jpmorgan.com, fernando.acero@ucl.ac.uk, sm2410@cam.ac.uk
Pseudocode No The paper contains flowcharts (Figure 1 and Figure 2) and mathematical formulations but no explicit pseudocode or algorithm blocks.
Open Source Code No The paper does not explicitly provide a statement or link for the open-source code of the methodology described.
Open Datasets No The paper mentions using 'real-world data' for IMP and 'synthetic data' for RR, with the generation process outlined in an extended version of the paper itself. No concrete access information, formal citations for established datasets, or links to publicly available datasets are provided.
Dataset Splits No The paper does not provide specific details regarding dataset splits for training, validation, or testing, such as percentages or sample counts.
Hardware Specification Yes Experiments were run on a 36 CPU, 72 GB RAM Linux machine.
Software Dependencies No The paper mentions the use of 'CPLEX commercial solver' but does not provide a specific version number or other software dependencies with versions.
Experiment Setup Yes Each experiment was performed with the following parameters: 500 scenarios (R = 500), 28 day horizon (T = 28), and 169 possible schedules (S = 169). For every implementation of Surrogate-MP, we deactivate the surrogate after the optimality gap is less than 5%, to focus on retrieval of optimality using the MIMP.