CombOptNet: Fit the Right NP-Hard Problem by Learning Integer Programming Constraints

Authors: Anselm Paulus, Michal Rolinek, Vit Musil, Brandon Amos, Georg Martius

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the potential of such layers with an extensive performance analysis on synthetic data and with a demonstration on a competitive computer vision keypoint matching benchmark. (...) First, we extensively analyze the performance on synthetic data. This includes the inverse optimization task of recovering an unknown set of constraints, and a KNAPSACK problem specified in plain text descriptions. Finally, we demonstrate the applicability to real-world tasks on a competitive computer vision keypoint matching benchmark. (...) We demonstrate the potential and flexibility of our method on four tasks.
Researcher Affiliation Collaboration 1Max-Planck-Institute for Intelligent Systems, T ubingen, Germany 2Masaryk University, Brno, Czechia 3Facebook AI Research, USA.
Pseudocode Yes Module 1 Comb Opt Net function FORWARDPASS(A, b, c) y := Solver(A, b, c) save y and A, b, c for backward pass return y function BACKWARDPASS(dy)...
Open Source Code No The paper does not provide an explicit statement or link for open-source code availability for the described methodology.
Open Datasets Yes For Random Constraints: "The dataset consists of 1 600 pairs (c, y ) for training and 1 000 for testing." For Weighted Set Covering: "the dataset consists of 1 600 pairs (c, y ) for training and 1 000 for testing." For KNAPSACK from Sentence Description: "The dataset contains 4 500 training and 500 test pairs (x, y )." For Deep Keypoint Matching: "We use the SPair-71k dataset (Min et al., 2019)... The dataset is split into 53 340 training pairs, 5 384 validation pairs and 12 234 pairs for testing."
Dataset Splits Yes The dataset is split into 53 340 training pairs, 5 384 validation pairs and 12 234 pairs for testing.
Hardware Specification No The paper does not specify any particular hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions using "GUROBI (Gurobi Optimization, 2019)" but does not provide a specific version number for it or any other software dependency like Python or relevant libraries.
Experiment Setup No The paper states: "Implementation details, a runtime analysis and additional results, such as ablations, other loss functions and more metrics, are provided in the Supplementary material." This indicates that specific setup details are not present in the main text.