ABSTRACT. With progress in AI and increasing computing power, new computer-aided design approaches have emerged where products are designed in a co-creative human--computer interaction. In architecture, interior design, and product design, computational methods such as evolutionary algorithms support the designer’s creative process by revealing populations of generated design solutions in a parametric design space. However, the benefits and shortcomings of such algorithms for designers are not yet fully understood. This paper reports on the in-depth, in-situ and longitudinal experiences of one industrial designer using interactive evolutionary algorithms in a creative product design task. The findings reveal inspirational impact to the designer by applying the tool, role of environment and long term impact to creativity in design processes, as well as some fixative impact. Also, it revealed a need for further studies on the system's role in design exploration and co-creative practice.

ABSTRACT. La muse is a joint project between artists and Machine Learning academic scholars. It aims at building pictorial compositions in order to provide sources of inspiration to assist painters in their process of creation. It relies on Artificial Intelligence, mainly based on various artificial neural networks, used for object recognition and style transfer. This article presents how, with minimal effort and without requiring extensive computational power La muse can take into account the visual universe of a painter, their artistic references, personal inspiration sources and preferred visual code books to create suggestions of painting subjects the human artist can then use as a source of inspiration for actual creation. Code developed in this project is Open Source and a free-to-use demonstration website is publicly accessible.

ABSTRACT. Although the narrative structure of common entertainment products like Hollywood movies or TV series is generally composed of a number of different plot lines combined into a single narrative discourse, efforts on computational modeling of story generation have to this point focused mostly on the construction of stories with a single plot line. The present paper explores an evolutionary solution to the task of building a story that combines more than one plot line into a single linear discourse. This requires a set of knowledge resources that capture the main features that influence the decisions involved, a representation suitable for evolutionary treatment for discourses with several plot lines, and a set of fitness functions based on metrics related to the quality of the resulting discourses are proposed. The proposed solution produces populations of stories with elaborate discourses that combine several subplots.

ABSTRACT. The measure of complexity is a core concept in computational approaches to aesthetics. Shannon's information theory provided an objective measure of complexity, which led to the emergence of various informational theories of aesthetics. However, entropy fails to take into account the spatial characteristics of 2D patterns; these characteristics are fundamental in addressing the aesthetic problem. We propose two empirically evaluated alternative measures of complexity, taking into account the spatial characteristics of 2D patterns and experimental studies on human aesthetic perception in the visual domain. The first model, spatial complexity, is based on the probabilistic spatial distribution of pixels of a 2D pattern. The second model is based on algorithmic information theory (Kolmogorov complexity) which is extended to estimate the complexity of 2D patterns. The spatial complexity measure presents a performance advantage over information-theoretic models, specifically in discriminating symmetries and the orientation in 2D, enabling more accurate measurement of complexity in relation to aesthetic evaluations of 2D patterns.

This paper examines whether the complexity measures that account for 2D structure conforms with aesthetic judgments. The results of this experiment showed that none of the three predictors had a substantial impact on ratings or rankings.

ABSTRACT. Differential evolution (DE) is widely used for global optimisation problems due to its simplicity and efficiency. L-SHADE is a state-of-the-art variant of DE algorithm that incorporates external archive, success-history-based parameter adaptation, and linear population size reduction. L-SHADE uses a current-to-pbest/1/bin strategy for mutation operator, while all individuals have the same probability to be selected. In this paper, we propose a novel L-SHADE algorithm, RWS-L-SHADE, based on a roulette wheel selection strategy so that better individuals have a higher priority and worse individuals are less likely to be selected. Our extensive experiments on the CEC-2017 benchmark functions and dimensionalities of 30, 50 and 100 indicate that RWS-L-SHADE outperforms L-SHADE.

ABSTRACT. Most of the Evolutionary Algorithms (EA) use a population of candidate solutions to explore the search space following specific rules during an iterative process. These algorithms are designed expecting a good balance between exploration and exploitation during the search process. Besides, the diversity of the population is crucial to properly explore the search space. This article introduces an improved version of the Differential Evolution (DE) algorithm which employs the moving average (MA) to determine when the population should diversify or intensify by using additional operators. The MA is one of the most used stock market indicators which provides recommendations for selling or buying stocks based on historical data. Here, the MA of the historical fitness and dimension-wise diversity is analyzed to determine if the DE continues operating normally or if it should diversify or intensify the search using additional operators.

ABSTRACT. Many frameworks and libraries are available for researchers working on optimization. However, the majority of them require programming knowledge, lack of a friendly user interface and cannot be run on different operating systems. WebGE is a new optimization tool which provides a web-based graphical user interface allowing any researcher to use Grammatical Evolution and Differential Evolution on symbolic regression problems. Besides, the fact that it can be deployed on any server as a web service also incorporating user authentication, makes it a versatile and portable tool that can be shared by multiple researchers. Finally, the modular software architecture allows to easily extend WebGE to other algorithms and types of problems.

ABSTRACT. Random Forest algorithm is a prediction model where a set of tree predictors are combined to construct an ensemble model. If a set of conditions are satisfied we can affirm that random forest avoids overfitting and converges. On the other hand, grammatical evolution, the popular variant of genetic programming where solutions are build following a grammar, has been succesfully applied to a plethora of different problems. Among them, symbolic regression are one of the hits of grammatical evolution. Although encoded in codons and decoded by a grammar, solutions in grammatical evolution are trees that represent mathematical expressions. In this paper we investigate the convenience of combining the best of both approaches and we propose Random Grammatical Evolution as an adaptation of Random Forest to symbolic regression problem. Random Grammatical Evolution builds a set of weak predictors that are combined on an ensemble model for prediction.

ABSTRACT. In the popular card game "The Resistance" two teams play against each other. Resistance fighters perform missions against an evil government while the government has sent out spies to infiltrate the resistance. Spies may sabotage the missions carried out by the resistance fighters. A main feature of the game is that the spies pose as resistance fighters. Successful spies trick the resistance in taking them on a mission and then try to sabotage the mission while still maintaining their cover. An important aspect of the game is that any kind of communication is allowed between game players as long as the communication is observable by all game players. We use an evolution strategy to evolve game strategies for resistance fighters as well as spies. In our experiment, game players only communicate via observable actions. Players are judged by how they behave and not by what they say. Resistance fighters observe the behavior of all game players and try to deduce who is a spy by maintaining a score that represents who is likely to be a spy. Players likely to be spies are not taken on a mission. Spies use probabilities for their behavior. We use co-evolution to evolve resistance fighters and spies. The master tournament and current individual vs ancestral opponents is used to show that evolutionary progress is being made.

ABSTRACT. Inspired by the recent 11th Global Trajectory Optimisation Competition, this paper presents the asteroid routing problem (ARP) as a realistic benchmark of algorithms for expensive bound-constrained black-box optimization in permutation space. Given a set of asteroids' orbits and a departure epoch, the goal of the ARP is to find the optimal sequence for visiting the asteroids, starting from Earth's orbit, in order to minimize both the cost, measured as the sum of the magnitude of velocity changes required to complete the trip, and the time, measured as the time elapsed from the departure epoch until visiting the last asteroid. We provide open-source code for generating instances of arbitrary sizes and evaluating solutions to the problem. As a preliminary analysis, we compare the results of two methods for expensive black-box optimization in permutation spaces, namely, CEGO, a Bayesian optimizer based on Gaussian processes, and UMM, an estimation-of-distribution algorithm based on probabilistic Mallows models. We investigate the best permutation representation for each algorithm, either rank-based or order-based. Moreover, we analyze the effect of providing a good initial solution, generated by a greedy nearest neighbor heuristic, on the performance of the algorithms. The results suggest directions for improvements in the algorithms being compared.

ABSTRACT. EvoCC framework is an open-source, free, and cross-platform framework implemented in Python which combines clustering, classification, and evolutionary computation methods. It optimizes the classification process by generating a classification model for each group generated by a clustering process where the clustering process is optimized by evolutionary optimization techniques. It includes the most well-known and recent nature-inspired metaheuristic optimization algorithms, well-known datasets, different fitness functions, and distance measures, and several well-known and highly-used classifiers. The aim is to provide the practitioners and researchers with a user-friendly and customizable implementation of classification-based nature-inspired optimization clustering algorithms that can be used by experienced and non-experienced users for the classification process in different domains. The current implementation of the framework includes eleven classification algorithms and five evaluation measures. It also utilizes the implementation of the EvoCluster framework which has ten metaheuristic optimizers, thirty datasets, five objective functions, more than twenty distance measures, and ten different ways for detecting the k value.

ABSTRACT. Mission critical services have stringent requirements in terms of reliability, energy-efficiency and performance. In this regard, optimisation approaches are required to maximise missions’ service time by minimising the consumption of energy without compromising performance like the throughput in scenarios with high density in terms of connected devices. EvoMCS is an evolutionary algorithm approach able to deter- mine the optimal configurations for IEEE 802.11 wireless networks in dense environments with multiple objectives such as minimisation of consumed energy and maximisation of the throughput in each device. EvoMCS is evaluated in wildfire scenarios, determining the optimal values for configuration parameters in 802.11n and 802.11ax technologies. The achieved results demonstrate that EvoMCS is able to provide optimal configuration values, that reduces in a factor of three the energy that is consumed by devices.