End-to-End Learning for Optimization via Constraint-Enforcing Approximators

Authors: Rares Cristian, Pavithra Harsha, Georgia Perakis, Brian L Quanz, Ioannis Spantidakis

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show experimental results on two end-to-end problems for a shortest path problem and a capacitated multiproduct newsvendor problem. In this section we present computational results that our Project Net method is effective in end-to-end learning.
Researcher Affiliation Collaboration Rares Cristian1, Pavithra Harsha2, Georgia Perakis1, Brian L Quanz2, Ioannis Spantidakis1 1 Massachusetts Institute of Technology 2 IBM T. J Watson Research Center
Pseudocode Yes Algorithm 1 Dykstra s Projection Method; Algorithm 2 Training Project Net; Algorithm 3 End-to-End Learning via Project Net
Open Source Code No The paper does not provide any specific repository links or explicit statements about the release of their own source code for the methodology described.
Open Datasets Yes We use the Warcraft II tile dataset (Guyomarch 2017) which was first introduced in (Vlastelica et al. 2020) to test their end-to-end approach for combinatorial problems. [...] Guyomarch, J. 2017. Warcraft ii open-source map editor. http://github.com/war2/war2edit.
Dataset Splits No The paper evaluates on 'testing data' but does not provide specific percentages, sample counts, or a clear methodology for dataset splitting into train/validation/test sets.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details, such as library or solver names with their version numbers, needed to replicate the experiment.
Experiment Setup Yes We also train a Project Net model with T0 = 5 iterations, and compare the objective value of its solution for iterations up to T1 = 35 on testing data. During training, we only perform T0 iterations of this mapping. In particular, once Project Net has been trained using T0 steps, we may then use some T1 > T0 steps when using it to learn forecasts. Algorithm 2 Training Project Net: Initialize matrix L, Initialize w0 = 0, zj+1 = wj ηL ui wj, wj+1 = πk(zj+1), Update L by any gradient method. Algorithm 3 End-to-End Learning via Project Net: Initialize θ at random. Update θ by any gradient method.