Pointer Networks Trained Better via Evolutionary Algorithms

TL;DR

The results of solving 1000-dimensional TSPs by training a PtrNet on the same dimensionality are reported, which strongly suggests that scaling up the training instances is in need to improve the performance of Ptr net on solving higher-dimensional COPs.

Abstract

Pointer Network (PtrNet) is a specific neural network for solving Combinatorial Optimization Problems (COPs). While PtrNets offer real-time feed-forward inference for complex COPs instances, its quality of the results tends to be less satisfactory. One possible reason is that such issue suffers from the lack of global search ability of the gradient descent, which is frequently employed in traditional PtrNet training methods including both supervised learning and reinforcement learning. To improve the performance of PtrNet, this paper delves deeply into the advantages of training PtrNet with Evolutionary Algorithms (EAs), which have been widely acknowledged for not easily getting trapped by local optima. Extensive empirical studies based on the Travelling Salesman Problem (TSP) have been conducted. Results demonstrate that PtrNet trained with EA can consistently perform much better inference results than eight state-of-the-art methods on various problem scales. Compared with gradient descent based PtrNet training methods, EA achieves up to 30.21\% improvement in quality of the solution with the same computational time. With this advantage, this paper is able to at the first time report the results of solving 1000-dimensional TSPs by training a PtrNet on the same dimensionality, which strongly suggests that scaling up the training instances is in need to improve the performance of PtrNet on solving higher-dimensional COPs.

Figures (7)

• Figure 1: Flowchart of PtrNet. The green arrows denote the output pointer and the red arrows denote the pointers disabled by the masking mechanism.
• Figure 2: The structure of the encoder module in PtrNet.
• Figure 3: The structure of the decoder module in PtrNet.
• Figure 4: A comparison of the convergence curves during the training process of other methods against PtrNet-EA. The three columns correspond to the training conducted on TSP100, TSP500, and TSP1000, respectively. And the three rows depict three groups of different compared algorithms. The vertical axis represents the Tour Length for the current batch on the Training set.
• Figure 5: 10x time budget for compared methods training on TSP1000 to reach the PtrNet-EA’s performance training in 500 minutes.