Neural TSP Solver with Progressive Distillation

Authors: Dongxiang Zhang, Ziyang Xiao, Yuan Wang, Mingli Song, Gang Chen

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that our approach establishes clear advantages over existing encoder-decoder models in terms of training effectiveness and solution quality.
Researcher Affiliation Academia 1 College of Computer Science and Technology, Zhejiang University 2 School of Business, Singapore University of Social Sciences
Pseudocode Yes The pseudo code of our model training based on progressive distillation is depicted in Algorithm 1.
Open Source Code Yes The datasets and source code are provided in the supplementary materials.
Open Datasets Yes For datasets, we follow previous learning-to-generate models to use uniform distribution in two-dimensional space [0, 1]2. In addition, we examine the performance in the Euclidean TSP instances in the TSPLIB repository3.
Dataset Splits No No explicit train/validation/test dataset splits (e.g., percentages or sample counts) are provided.
Hardware Specification Yes All the experiments are conducted on a single GPU (NVIDIA Tesla V100 with 32GB memory).
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries used.
Experiment Setup Yes As to hyper-parameter setting, we set N = 100, C = 60, r = 0.4, β = 0.1 and T = 2000 for the training framework based on progressive distillation. The Transformer has 6-layers encoder and 1-layer decoder, both with 128 hidden units in each layer.For batch size, we set 512 for TSP150, 128 for TSP300 and 32 for TSP500.