Traveling Salesman Problems (TSPs) have been a long-lasting interesting challenge to researchers in different areas. The difficulty of such problems scales up further when multiple objectives are considered concurrently. Plenty of work in evolutionary algorithms has been introduced to solve multi-objective TSPs with promising results, and the work in deep learning and reinforcement learning has been surging. This paper introduces a multi-objective deep graph pointer network-based reinforcement learning (MODGRL) algorithm for multi-objective TSPs. The MODGRL improves an earlier multi-objective deep reinforcement learning algorithm, called DRL-MOA, by utilizing a graph pointer network to learn the graphical structures of TSPs. Such improvements allow MODGRL to be trained on a small-scale TSP, but can find optimal solutions for large scale TSPs. NSGA-II, MOEA/D and SPEA2 are selected to compare with MODGRL and DRL-MOA. Hypervolume, spread and coverage over Pareto front (CPF) quality indicators were selected to assess the algorithms’ performance. In terms of the hypervolume indicator that represents the convergence and diversity of Pareto-frontiers, MODGRL outperformed all the competitors on the three well-known benchmark problems. Such findings proved that MODGRL, with the improved graph pointer network, indeed performed better, measured by the hypervolume indicator, than DRL-MOA and the three other evolutionary algorithms. MODGRL and DRL-MOA were comparable in the leading group, measured by the spread indicator. Although MODGRL performed better than DRL-MOA, both of them were just average regarding the evenness and diversity measured by the CPF indicator. Such findings remind that different performance indicators measure Pareto-frontiers from different perspectives. Choosing a well-accepted and suitable performance indicator to one’s experimental design is very critical, and may affect the conclusions. Three evolutionary algorithms were also experimented on with extra iterations, to validate whether extra iterations affected the performance. The results show that NSGA-II and SPEA2 were greatly improved measured by the Spread and CPF indicators. Such findings raise fairness concerns on algorithm comparisons using different fixed stopping criteria for different algorithms, which appeared in the DRL-MOA work and many others. Through these lessons, we concluded that MODGRL indeed performed better than DRL-MOA in terms of hypervolumne, and we also urge researchers on fair experimental designs and comparisons, in order to derive scientifically sound conclusions.

Short term memory that records the current population has been an inherent component of Evolutionary Algorithms (EAs). As hardware technologies advance currently, inexpensive memory with massive capacities could become a performance boost to EAs. This paper introduces a Long Term Memory Assistance (LTMA) that records the entire search history of an evolutionary process. With LTMA, individuals already visited (i.e., duplicate solutions) do not need to be re-evaluated, and thus, resources originally designated to fitness evaluations could be reallocated to continue search space exploration or exploitation. Three sets of experiments were conducted to prove the superiority of LTMA. In the first experiment, it was shown that LTMA recorded at least $50\%$ more duplicate individuals than a short term memory. In the second experiment, ABC and jDElscop were applied to the CEC-2015 benchmark functions. By avoiding fitness re-evaluation, LTMA improved execution time of the most time consuming problems $F03$ and $F05$ between 7% and 28% and 7% and 16%, respectively. In the third experiment, a hard real-world problem for determining soil models’ parameters, LTMA improved execution time between 26% and 69%. Finally, LTMA was implemented under a generalized and extendable open source system, called EARS. Any EA researcher could apply LTMA to a variety of optimization problems and evolutionary algorithms, either existing or new ones, in a uniform way.

This paper describes a research work on Semantic Inference, which can be regarded as an extension of Grammar Inference. The main task of Grammar Inference is to induce a grammatical structure from a set of positive samples (programs), which can sometimes also be accompanied by a set of negative samples. Successfully applying Grammar Inference can result only in identifying the correct syntax of a language. With the Semantic Inference a further step is realised, namely, towards inducing language semantics. When syntax and semantics can be inferred, a complete compiler/interpreter can be generated solely from samples. In this work Evolutionary Computation was employed to explore and exploit the enormous search space that appears in Semantic Inference. For the purpose of this research work the tool LISA.SI has been developed on the top of the compiler/interpreter generator tool LISA. The first results are encouraging, since we were able to infer the semantics only from samples and their associated meanings for several simple languages, including the Robot language.

Multi-Objective Evolutionary Algorithms (MOEAs) have been applied successfully for solving real-world multi-objective problems. Their success can depend highly on the configuration of their control parameters. Different tuning methods have been proposed in order to solve this problem. Tuning can be performed on a set of problem instances in order to obtain robust control parameters. However, for real-world problems, the set of problem instances at our disposal usually are not very plentiful. This raises the question: What is a sufficient number of problems used in the tuning process to obtain robust enough parameters? To answer this question, a novel method called MOCRS-Tuning was applied on different sized problem sets for the real-world integration and test order problem. The configurations obtained by the tuning process were compared on all the used problem instances. The results show that tuning greatly improves the algorithms’ performance and that a bigger subset used for tuning does not guarantee better results. This indicates that it is possible to obtain robust control parameters with a small subset of problem instances, which also substantially reduces the time required for tuning.