Learning Solution-Aware Transformers for Efficiently Solving Quadratic Assignment Problem

Authors: Zhentao Tan, Yadong Mu

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our model s effectiveness is validated through extensive experiments on self-generated QAP instances of varying sizes and the QAPLIB benchmark.
Researcher Affiliation Academia 1Center for Data Science, Peking University, China. 2Wangxuan Institute of Computer Technology, Peking University, China.
Pseudocode Yes At first, we initialize the facility nodes with vectors randomly sampled from a one-hot vector pool with a dimension of Ninit = 128. Details are shown below in Python code. 1 f_init_emb = torch.zeros(size=(batch_size, node_cnt, N_{init})).to(device)...
Open Source Code Yes The corresponding code and other resources are released at https://github.com/PKUTAN/SAWT.
Open Datasets Yes QAPLIB (Burkard et al., 1997) benchmarks showcase the model s effectiveness and robust generalization to instances of varying sizes.
Dataset Splits Yes For all experiments, we use a training set of up to 5120 instances and evaluate results on a test set of 256 different instances from the same distribution. The gap, averaged over 256 validation instances across 400 search steps during training...
Hardware Specification Yes All experiments are executed on a single NVIDIA 3080Ti GPU (12GB) and a 12th Gen Intel(R) Core(TM) i5-12600KF 3.69 GHz CPU.
Software Dependencies No Our model SAWT is implemented using PyTorch (Paszke et al., 2019). While PyTorch is mentioned, a specific version number is not provided, nor are other software dependencies with version numbers.
Experiment Setup Yes We train the model for 200 epochs with a batch size of 512 and an episode length of 400 for all tasks. Table 7 below summarizes the hyper-parameter settings: HYPERPARAMETER QAP10 QAP20 QAP50 QAP100 demb 64 64 64 64 dhidden 64 64 64 64 L 3 2 3 3 lr 10 3 10 3 10 3 10 3