Test-Time Augmentation for Traveling Salesperson Problem
This addresses combinatorial optimization problems for researchers and practitioners, representing an incremental improvement through augmentation of existing methods.
The paper tackles the Traveling Salesperson Problem by proposing Test-Time Augmentation (TTA), which interprets permutation of node indices as an augmentation scheme to improve solution quality. The results show the method obtains shorter solutions than latest models and increases the probability of finding solutions closer to exact solutions with larger augmentation sizes.
We propose Test-Time Augmentation (TTA) as an effective technique for addressing combinatorial optimization problems, including the Traveling Salesperson Problem. In general, deep learning models possessing the property of invariance, where the output is uniquely determined regardless of the node indices, have been proposed to learn graph structures efficiently. In contrast, we interpret the permutation of node indices, which exchanges the elements of the distance matrix, as a TTA scheme. The results demonstrate that our method is capable of obtaining shorter solutions than the latest models. Furthermore, we show that the probability of finding a solution closer to an exact solution increases depending on the augmentation size.