Controlling Continuous Relaxation for Combinatorial Optimization
Authors: Yuma Ichikawa
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that CRA significantly enhances the performance of UL-based solvers, outperforming existing UL-based solvers and greedy algorithms in complex CO problems. |
| Researcher Affiliation | Collaboration | Yuma Ichikawa Fujitsu Limited, Kanagawa, Japan Department of Basic Science, University of Tokyo |
| Pseudocode | No | The paper includes mathematical formulations and descriptions of algorithms, but no structured pseudocode or algorithm blocks are explicitly provided. |
| Open Source Code | Yes | The code is available at https://github.com/Yuma-Ichikawa/CRA4CO. |
| Open Datasets | Yes | We evaluate our method using the MIS benchmark dataset from recent studies [Goshvadi et al., 2023, Qiu et al., 2022], which includes graphs from SATLIB [Hoos and Stützle, 2000] and Erd os Rényi graphs (ERGs) of varying sizes. |
| Dataset Splits | No | The paper mentions early stopping which monitors loss, but it does not specify explicit training/validation/test dataset splits or a dedicated validation set for monitoring performance. |
| Hardware Specification | No | The paper mentions 'Pytorch GPU' but does not provide specific details such as GPU model, number of GPUs, CPU specifications, or memory, which are necessary for hardware reproducibility. |
| Software Dependencies | No | The paper mentions software components like 'Deep Graph Library', 'Adam W optimizer', 'CVXOPT solver', and 'CVXPY' but does not provide specific version numbers for any of them. |
| Experiment Setup | Yes | We use the Adam W [Kingma and Ba, 2014] optimizer with a learning rate of η = 10 4 and weight decay of 10 2. The GNNs are trained for up to 5 104 epochs with early stopping... We set the initial scheduling value to γ(0) = 20 for the MIS and matching problems, and we set γ(0) = 6 for the Max Cut problems with the scheduling rate ε = 10 3 and curve rate α = 2 in Eq. (3.2). |