Inverse Optimization via Learning Feasible Regions

Authors: Ke Ren, Peyman Mohajerin Esfahani, Angelos Georghiou

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Synthetic experiments and two power system applications, including comparisons with state-of-the-art approaches, showcase and validate the proposed approach.
Researcher Affiliation Collaboration 1Amazon 2University of Toronto and Delft University of Technology 3University of Cyprus. Correspondence to: Ke Ren <EMAIL>.
Pseudocode Yes Algorithm 1 Gradient descent based algorithm for (10) using predictability loss
Open Source Code No No explicit statement or link indicating the release of source code for the methodology described in this paper was found.
Open Datasets Yes We apply our methods on an instance of a power system described in (Bampou & Kuhn, 2011). ... We further investigate the performance of the proposed approach using the larger IEEE 14-bus system (Leon et al., 2020).
Dataset Splits Yes We generate Ntrain = 100 data points for training and Ntest = 200 data points for testing, by generating signals uniformly at random from scost 1 [0.2, 1], scost 2 [0.2, 0.5], scost 3 [1, 2] and sdemand 1 [0.3, 1.5], sdemand 2 [0.36, 1.8], sdemand 3 [0.42, 2.1], sdemand 4 [0.48, 2.4] and sdemand 5 [0.54, 2.7], and solving problem (14) to obtain pairs {si, xi}N=100 i=1 .
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments. It mentions Gurobi, which is a solver, not hardware.
Software Dependencies Yes We use the hypothesis (7) where Z is the unit simplex of dimension p {3, 6, 9} and solve problems (10) using the non-convex quadratic solver of Gurobi v11.0.3 with a time limit of 1800 seconds.
Experiment Setup Yes We use the hypothesis (7) where Z is the unit simplex of dimension p {3, 6, 9} using the Adaptive Smoothing Algorithm 2 with a limit of 3000 iterations. ... The initial values are ϵ1 = ϵ2 = 1, and every time the change in the values of PN i=1 γs1 2 and PN i=1 γs2 2 are less than 0.01/10(log2(ϵ1)+1), we multiply the parameters ϵ1 and ϵ2 by 2.