No abstract

This work assesses the efficacy of evolutionary algorithms (EAs) using an intuitive Multi-Dimensional Scaling (MDS) visualisation of the evolution of a population. We propose the use of Landmark Multi-Dimensional Scaling (LMDS) to overcome computational challenges inherent to visualising many-objective and complex problems with MDS. For the benchmark problems we tested, LMDS is akin to MDS visually, whilst requiring less than 1% of the time and memory necessary to produce an MDS visualisation of the same objective space solutions, leading to the possibility of online visualisations for multiand many-objective optimisation evaluation. Using multi-and many-objective problems from the DTLZ and WFG benchmark test suites, we analyse how Landmark MDS visualisations can offer far greater insight into algorithm performance than using traditional algorithm performance metrics such as hypervolume alone, and can be used to complement explicit performance metrics. Ultimately, this visualisation allows visual identification of problem features and assists the decision maker in making intuitive recommendations for algorithm parameters/operators for creating and testing better EAs to solve multi-and manyobjective problems.

The comprehension of the Evolutionary Algorithm (EA) search process is often eluded by challenges of transparency inherent to black-box EAs, thus affecting algorithm enhancement and hyperparameter optimisation. In this work, we develop algorithm insight by introducing the Population Dynamics Plot (PopDP). PopDP is a novel and intuitive visualisation capable of visualising the population of solutions, the parent-offspring lineage, solution perturbation operators, and the search process journey. We apply PopDP to NSGA-II to demonstrate the insight attained and the effectiveness of PopDP for visualising algorithm search on a series of discrete dualand many-objective knapsack problems of different complexities, and our results demonstrate that the method can be used to produce a visualisation in which the lineage of solutions can be clearly seen. We also consider the efficacy of the proposed explainable visualisation against emerging approaches to benchmarking explainable AI methods and consider the accessibility of the resulting visualisations.

Evolutionary algorithms are widely used to solve optimisation problems. However, challenges of transparency arise in both visualising the processes of an optimiser operating through a problem and understanding the problem features produced from many-objective problems, where comprehending four or more spatial dimensions is difficult. This work considers the visualisation of a population as an optimisation process executes. We have adapted an existing visualisation technique to multi-and many-objective problem data, enabling a user to visualise the EA processes and identify specific problem characteristics and thus providing a greater understanding of the problem landscape. This is particularly valuable if the problem landscape is unknown, contains unknown features or is a many-objective problem. We have shown how using this framework is effective on a suite of multi-and many-objective benchmark test problems, optimising them with NSGA-II and NSGA-III.

Contemporary maritime operations such as shipping are a vital component constituting global trade and defence. The evolution towards maritime autonomous systems, often providing significant benefits (e.g., cost, physical safety), requires the utilisation of artificial intelligence (AI) to automate the functions of a conventional crew. However, unsecured AI systems can be plagued with vulnerabilities naturally inherent within complex AI models. The adversarial AI threat, primarily only evaluated in a laboratory environment, increases the likelihood of strategic adversarial exploitation and attacks on mission-critical AI, including maritime autonomous systems. This work evaluates AI threats to maritime autonomous systems in situ. The results show that multiple attacks can be used against real-world maritime autonomous systems with a range of lethality. However, the effects of AI attacks vary in a dynamic and complex environment from that proposed in lower entropy laboratory environments. We propose a set of adversarial test examples and demonstrate their use, specifically in the marine environment. The results of this paper highlight security risks and deliver a set of principles to mitigate threats to AI, throughout the AI lifecycle, in an evolving threat landscape.