ABSTRACT. Seismic history matching (SHM) is a challenging industry problem for optimising the production from subsurface reservoirs. The reservoir models are optimised by matching their simulated data to the observed production and time-lapse (4D) seismic data. These data are often sparse and contain errors, so the assimilated models for real reservoirs typically have poorly constrained parameters and are non-unique. To understand this problem at a deeper level, we explore the topology of their search space using local optima networks (LONs), a powerful tool for characterising fitness landscapes. We extend the use of LONs to the real-world, computationally expensive, SHM problem. We found that the LONs were different for objective functions based on both production data and time-lapse reservoir maps, and for each dimensionality. Objective functions based on well pressures and oil saturation maps had the highest success rates, but the number of optima and funnels increased with dimensionality for all objective functions. In contrast, the success rate and strength of the global optima decreased significantly with increasing dimensionality. Our work goes some ways to explaining the results of real SHM problems in industry, and demonstrates the value of fitness landscape analysis, using LONs, for computationally expensive problems.

ABSTRACT. The proliferation of metamorphic malware has recently gained a lot of research interest. This is because of their ability to transform their program codes stochastically. Several detectors are unable to detect this malware family because of how quickly they obfuscate their code. Machine learning (ML) models have also been shown to not be robust to these attacks due to the insufficient data to train these models resulting from the constant code mutation of metamorphic malware. Although recent studies have shown how to generate samples of metamorphic malware to serve as training data, this process can be computationally expensive. One way to improve the performance of these ML models is to transfer learning from other fields which have robust models such as what has been done with the transfer of learning from computer vision and image processing to improve malware detection. In this work, we introduce an evolutionary based transfer learning approach that uses evolved mutants of malware generated using a traditional Evolutionary Algorithm (EA) as well as models from Natural Language Processing (NLP) text classification to improve the classification of metamorphic malware. Our preliminary results demonstrate that using NLP models can improve the detection of metamorphic malware in some instances.

ABSTRACT. This work investigates the generation of fractal pattern structures representing the phenotype of an organism, using the Barnsley fern as an example. To do so, a standard Genetic Algorithm is implemented to evolve the parameters of the Iterated Function System and select the resulting fractal structures, each representing a generated phenotype, using a box-counting dimension as a fitness metric. In this way, realistic fern structures are evolved over a few tens of generations. The algorithm is further extended to test its potential to generate other natural fractals.

ABSTRACT. Automated per-instance configuration and selection of algorithms are gaining significant moments in evolutionary computation in recent years. Two crucial, sometimes implicit, ingredients for these AutoML methods are 1) feature-based representations of the problem instances and 2) performance prediction methods that take these features as input to estimate how well a specific algorithm instance will perform on a given problem instance. Non-surprisingly, common ML models fail to make predictions for instances whose feature-based representation is underrepresented or not covered in the training data, resulting in poor generalization ability of the models for problems not seen during training.

In this work, we study leave-one-problem-out performance prediction. We analyze whether standard random forest (RF) model predictions can be improved by calibrating it with a weighted average of performance values obtained by the algorithm on problem instances that are sufficiently close to the problem for which a performance prediction is sought, measured by cosine similarity in feature space.

While our RF+clust approach obtains more accurate performance prediction for several problems, its predictive power crucially depends on the chosen similarity threshold as well as on the feature portfolio for which the cosine similarity is measured, thereby opening a new angle for feature selection in a zero-shot learning setting, as LOPO is termed in machine learning.

ABSTRACT. Quality-diversity (QD) methods such as MAP-Elites have been demonstrated to be useful in the domain of combinatorial optimisation due to their ability to generate a large set of solutions to a single-objective problem that are diverse with respect to user-defined features of interest. However, filling a MAP-Elites container with solutions can require careful design of operators to ensure complete exploration of the feature-space. Working in the domain of urban logistics, we propose two methods to increase exploration. Firstly, we exploit multiple decodings of the same genome which can generate different offspring from the same parent solution. Secondly, we make use of a multiple mutation operators to generate offspring from a parent, using a multi-armed bandit algorithm to adaptively select the best operator during the search. Our results on a set of 48 instances show that both the number of solutions within the container and the qd score of the container (indicating quality) can be significantly increased compared to the standard MAP-Elites approach.

ABSTRACT. Symmetry is a universal concept, its unique importance has made it a topic of research across many different fields. It is often considered as a constant where higher levels of symmetry are preferred in the judgement of faces and even the initial state of the universe is thought to have been in pure symmetry. The same is true in the judgement of auto-generated art, with symmetry often used alongside complexity to generate aesthetically pleasing images; however, these are two of many different aspects contributing to aesthetic judgement, each one of these aspects is also influenced by other aspects, for example, art expertise. These intricacies cause multiple problems such as making it difficult to describe aesthetic preferences and to auto-generate artwork using a high number of these aspects. In this paper, a gamified approach is presented which is used to elicit the preferences of symmetry levels for individuals and further understand how symmetry can be utilised within the context of automatically generating artwork. The gamified approach is implemented within an experiment with participants aged between 13 and 60, providing evidence that symmetry should be kept consistent within an evolutionary art context.

ABSTRACT. The impact of internet use on the environment is a contradictory topic. While it can help reduce carbon emissions, with smart grids or the automation of services and resources, it can also increase e-waste that end up affecting the environment. To draw attention to the impact of energy consumption on the environment, we proposed and developed a computational artifact that unites the areas of Data Aesthetics and Interaction Design. The artifact, displayed in an interactive installation, was divided into three panels: (i) the left panel, which represents the countries—from the European Union (EU)—with the lowest energy consumption impact on the environment; (ii) the central panel, which use swarming boids to represent the internet use at the installation site and its impact; and (iii) the right panel, which represents the EU countries with the highest energy impact on the environment. The arrangement of the three panels in a single interactive installation aims to establish a visual connection between the energy consumption in the EU and the energy consumption in the installation’s site and to promote awareness of its impact on the environment.

ABSTRACT. The dissemination of open-source text-to-image generative models and the increasing quality of their output has led to a growth in interest in the field. In this paper, we introduce an evolutionary approach to prompt generation where users begin by creating a blueprint for what might be a candidate prompt and then initiate an evolutionary process to interactively explore the space of prompts encoded by the initial blueprint and according to their preferences. Our work is a step towards a more dynamic and interactive way to generate prompts that lead to a wide variety of visual outputs, with which users can easily obtain prompts that match their goals.

ABSTRACT. In recent years Generative Machine Learning systems have advanced significantly. A current wave of generative systems use text prompts to create complex imagery, video, even 3D datasets. The creators of these systems claim a revolution in bringing creativity and art to anyone who can type a prompt. In this position paper, we question the basis for these claims, dividing our analysis into three areas: the limitations of linguistic descriptions, implications of the dataset, and lastly, matters of materiality and embodiment. We conclude with an analysis of the creative possibilities enabled by prompt-based systems, asking if they can be considered a new artistic medium.

ABSTRACT. Dynamic Multi-workflow Scheduling (DMWS) in cloud computing is a well-known combinatorial optimisation problem. It is a great challenge to tackle this problem by scheduling multiple workflows submitted at different times and meet user-defined quality of service objectives. Scheduling with user-defined budget constraints is becoming increasingly important due to cloud dynamics associated with on-demand provisioning, instance types and pricing. To address the Budget-Constrained Dynamic Multi-workflow Scheduling (BC-DMWS) problem, a novel Cooperative Coevolution Genetic Programming (CCGP) approach is proposed. Two heuristic rules, namely VM Selection/Creation Rule (VMR) and Budget Alert Rule (BAR), are learned automatically by CCGP. VMR is used to allocate ready tasks to either existing or newly rented VM instances, while BAR makes decisions to downgrade VM instances so as to meet the budget constraint. Experiments show significant performance and success rate improvement compared to several state-of-the-art algorithms.

ABSTRACT. This paper investigated the impacts of multi-objectivization on solving combinatorial single-objective $NK$-landscape problems with multiple funnel structures. Multi-objectivization re-formulates a single-objective target problem into a multi-objective problem with a helper problem in order to suppress the difficulty of the target problem. This paper analyzed the connectivity of two funnels respectively involving global optima in the target and the helper $NK$-landscape problems via the Pareto local optimal solutions in the multi-objectivized problem. Experimental results showed that multi-objectivization connects the two funnels with global optima of the target and the helper problems as a single bridging domain consisting of the Pareto local optimal solutions. Also, this paper proposed an algorithm named the multi-objectivized local search (MOLS) that searched for the global optimum of the target problem from the global optimum of an artificially generated helper problem via the Pareto local optimal solutions. Experimental results showed that the proposed MOLS achieved a higher success rate of the target single-objective optimization than iterative local search algorithms on target $NK$-landscape problems with multiple funnels.

ABSTRACT. This paper adapts a graph-based analysis and visualisation tool, search trajectory networks (STNs) to multi-objective combinatorial optimisation. We formally define multi-objective STNs and apply them to study the dynamics of two state-of-the-art multi-objective evolutionary algorithms: MOEA/D and NSGA2. In terms of benchmark, we consider two- and three-objective rmnk s for constructing multi-objective multi-modal landscapes with objective correlation. We find that STN metrics and visualisation offer valuable insights into both problem structure and algorithm performance. Most previous visual tools in multi-objective optimisation consider the objective space only. Instead, our newly proposed tool asses algorithm behaviour in the design and objective spaces simultaneously.

ABSTRACT. The interest in AutoML search spaces has given rise to a plethora of studies conceived to better understand the characteristics of these spaces, and the use of exploratory landscape analysis is among the most commonly explored techniques. However, in contrast with other classical optimization problems, defining the landscape may be as tough as characterizing it. This is because the concept of solution neighborhood is far from clear, as the spaces have a high number of conditional hyperparameters and a somehow hierarchical structure. This paper looks at the impact of different solution representations and distance metrics on the definition of these spaces, and how they affect exploratory landscape analysis metrics. We conclude that these metrics are not able to deal with structured, complex spaces such as the AutoML ones, and problem-related metrics might be the way to leverage this landscape complexity.