Offline Learning for Selection Hyper-heuristics with Elman Networks

Abstract

Abstract. Offline selection hyper-heuristics are machine learning methods that are trained on heuristic selections to create an algorithm that is tuned for a particular problem domain. In this work, a simple selection hyper-heuristic is executed on a number of computationally hard benchmark optimisation problems, and the resulting sequences of low level heuristic selections and objective function values are used to construct an offline learning database. An Elman network is trained on sequences of heuristic selections chosen from the offline database and the network's ability to learn and generalise from these sequences is evaluated. The networks are trained using a leave-one-out cross validation methodology and the sequences of heuristic selections they produce are tested on benchmark problems drawn from the HyFlex set. The results demonstrate that the Elman network is capable of intra-domain learning and generalisation with 99% confidence and produces better results than the training sequences in many cases. When the network was trained using an interdomain training set, the Elman network did not exhibit generalisation indicating that inter-domain generalisation is a harder problem and that strategies learned on one domain cannot necessarily be transferred to another.

A Methodology to Determine the Subset of Heuristics for Hyperheuristics through Metalearning for Solving Graph Coloring and Capacitated Vehicle Routing Problems

“…Usually, the offline selection hyperheuristics belong to machine learning methods, which are trained to create a tuned methodology for a problem domain [3]. Yates and Keedwell [19] demonstrated that subsequences of heuristics were found in the offline learning database that is effective for some problem domains.…”

“…e aim of this paper is not to present a survey on heuristics or hyperheuristics; our proposal is slightly different. Our proposal considers some vital aspects of the research, including the ones from Yates and Keedwell [19]. We took the offline hyperheuristic approach from Soria-Alcaraz et al [20] and the statistical approach to selecting a pool heuristic from Kanda et al [5].…”

A review on the self and dual interactions between machine learning and optimisation

“…In [127], the author used NNs to learn the hidden patterns between problem states and the promising low-level heuristics, in which the problem states were represented by constraint density and tightness. The approach in [194] learned recurrent neural networks (RNNs) [78] to predict next suitable heuristic based on a set of promising heuristic sequences according to the final log return. The trained RNNs can be used later to generate a sequence of heuristics on an unseen problem.…”