The implementation of a pointer network model for traveling salesman problem on a xilinx PYNQ board

Abstract

In this paper, a pointer network model for traveling salesman problem (TSP) was implemented on a Xilinx PYNQ board which supports Python and Jupyter notebook and is equipped with ZYNQ SOC. We implement a pointer network model for solving TSP with Python and Theano firstly, then train the model on a GPU platform, and eventually deploy the model on a PYNQ board. Unlike traditional neural network implementation, hardware libraries on PYNQ (Overlays) are used to accelerate the pointer network model application. The experimental results show that the pointer network model for TSP can be deployed on the embedded system successfully and achieve good performance.