ABSTRACT. When generating output from generative deep neural networks for creative applications it is common to explore multiple sampling approaches. This sampling stage is a crucial step, as choosing suitable sampling parameters can make or break the realism and perceived creative merit of the output. The process of selecting the correct sampling parameters is often task-specific and under-reported in many publications, which can make the reproducibility of the results challenging. We explore some of the most common sampling techniques in the context of generating human body movement, specifically dance movement, and attempt to shine a light on their advantages and limitations. This work presents a Mixture Density Recurrent Neural Network trained on a dataset of improvised dance motion capture data from which it is possible to generate novel movement sequences. We outline several common sampling strategies for Mixture Density Recurrent Neural Networks (MDRNNs) and examine these strategies systematically to further understand the effects of sampling parameters on motion generation. This analysis provides evidence that the choice of sampling strategy significantly affects the output of the model and supports the use of this model in creative applications. Building an understanding of the relationship between sampling parameters and creative machine-learning outputs could aid when deciding between different approaches in generation of dance motion and other creative applications.

ABSTRACT. Automated computer generation of aesthetically pleasing artwork has been the subject of research for several decades. The unsolved problem of interest is how to automatically please any audience without too much involvement of the said audience in the process of creation. Two dimensional pictures have received a lot of attention, however 3D artwork has remained relatively unexplored. This paper introduces the Axial Generation Process (AGP), a versatile generation algorithm that can be employed to create both 2D and 3D items within the Concretism art style. The evaluation of items generated using the process using a set of formal aesthetic measures, shows the process to be capable of generating visually varied items which generally exhibit a diverse range of values across the measures used, in both two and three dimensions.

ABSTRACT. This paper addresses the computational notion of aesthet- ics in the framework of multi-state two-dimensional cellular automata (2D CA). The measure of complexity is a core concept in computational approaches to aesthetics. Shannon's information theory provided an ob- jective 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, in general, and of CA-generated patterns, in particular. We propose two empirically evaluated alternative measures of complexity, taking into account the spatial characteristics of 2D patterns and experimental studies on hu- man aesthetic perception in the visual domain. The first model, spatial complexity, is based on the probabilistic spatial distribution of neighbouring cells over the lattice of a multi-state 2D cellular automaton. The second model is based on algorithmic information theory (Kolmogorov complexity) which is extended to estimate the complexity of 2D pat- terns. The spatial complexity measure presents performance advantage over information-theoretic models, specifically in discriminating symme- tries and the orientation in CA-generated patterns, enabling more accu- rate measurement of complexity in relation to aesthetic evaluations of 2D patterns. The results of experimentation demonstrate the presence of correlation between spatial complexity measures and aesthetic judge- ments of experimental stimuli. The same results were obtained for the estimations of Kolmogorov complexity of experimental stimuli.

ABSTRACT. This paper introduces RL Brush, a level-editing tool for tile-based games designed for mixed-initiative co-creation. The tool uses reinforcement-learning-based models to augment manual human level-design through the addition of AI-generated suggestions. Here, we apply RL Brush to designing levels for the classic puzzle game Sokoban. We put the tool online and tested it with 39 different sessions. The results show that users using the AI suggestions stay around longer and their created levels on average are more playable and more complex than without.

ABSTRACT. This paper presents a study conducted in naturalistic setting with data collected from an interactive art installation. The audience is challenged in a Turing Test for artistic creativity involving recognising human-made versus AI-generated drawing strokes. In most cases, people were able to differentiate human-made strokes above chance. An analysis conducted on the images at the pixel level shows a significant difference between the symmetry of the AI-generated strokes and the human-made ones. However we argue that this feature alone was not key for the dif- ferentiation. Further behavioural analysis indicates that people judging more quickly were able to differentiate human-made strokes significantly better than the slower ones. We point to theories of embodiment as a possible explanation of our results.

ABSTRACT. Weight initialization is critical in being able to successfully train artificial neural networks (ANNs), and even more so for recurrent neural networks (RNNs) which can easily suffer from vanishing and exploding gradients. In neuroevolution, where evolutionary algorithms are applied to neural architecture search, weights typically need to be initialized at three different times: when initial genomes (ANN architectures) are created at the beginning of the search, when offspring genomes are generated by crossover, and when new nodes or edges are created during mutation. This work explores the difference between using Xavier, Kaiming, and uniform random weight initialization methods, as well as novel Lamarckian weight inheritance methods for initializing new weights during crossover and mutation operations. These are examined using the Evolutionary eXploration of Augmenting Memory Models (EXAMM) neuroevolution algorithm, which is capable of evolving RNNs with a variety of modern memory cells (e.g., LSTM, GRU, MGU, UGRNN and Delta-RNN cells) as well as recurrent connections with varying time skips through a high performance island based distributed evolutionary algorithm. Results show that with statistical significance, utilizing the Lamarckian strategies outperforms Kaiming, Xavier and uniform random weight initialization, can speed neuroevolution by requiring less backpropagation epochs to be evaluated for each generated RNN and that the neuroevolutionary process provides further benefits to neural network weight optimization.

ABSTRACT. Neuroevolution commonly uses speciation strategies to better explore the search space of neural network architectures. One such speciation strategy is the use of islands, which are also popular in improving performance and convergence of distributed evolutionary algorithms. However, islands may experience stagnation, which prevents their convergence towards better solutions and result in wasted computation. In this paper, we propose utilizing an island extinction and repopulation mechanism to avoid premature convergence. We explore the performance of this method with the Evolutionary eXploration of Augmenting Memory Models (EXAMM), an asynchronous island based neuroevolution algorithm which progressively evolves recurrent neural networks (RNNs). In this strategy, all members of the worst performing island are erased periodically and repopulated with mutated versions of the global best genome. This island based strategy is additionally compared to NEAT’s (NeuroEvolution of Augmenting Topologies) speciation strategy. Experiments were performed using two different real world time series datasets (coal-fired power plant and aviation flight data). The results show that in addition to being more scalable, with statistical significance, this island extinction and repopulation strategy evolves better global best genomes than both EXAMM's original island based strategy and NEAT’s speciation strategy. The extinction and repopulation strategy is easy to implement, and can be generically applied to other neuroevolution algorithms.

ABSTRACT. Recent studies have shown that Deep Leaning models are susceptible to adversarial examples, which are data, in general images, intentionally modified to fool a machine learning classifier. In this paper, we present a multi-objective nested evolutionary algorithm to generate universal unrestricted adversarial examples in a black-box scenario. The unrestricted attacks are performed through the application of well known image filters that are available in several image processing libraries, modern cameras and mobile applications. The multi-objective optimization takes into account not only the attack success rate but also the detection rate. Experimental results showed that this approach is able to create a sequence of filters capable of generating very effective and undetectable attacks.

ABSTRACT. The ability of evolutionary processes to innovate and scale up over long periods of time, observed in nature, remains a central mystery in evolutionary biology, and a challenge for algorithm designers to emulate and explain in evolutionary computation (EC). The Major Transitions in Evolution is a compelling theory that explains evolvability through a multi-scale process whereby individuality (and hence selection and variation) is continually revised by the formation of associations, a process still not fully explored in EC. Deep Optimisation (DO) is a new type of model-building optimization algorithm (MBOA) that exploits deep learning methods to enable multi-scale optimization. DO uses an autoencoder model to induce a multi-level representation of solutions, capturing the relationships between the lower-level units that contribute to the quality of a solution. Variation and selection are then performed at these representations, causing a simultaneous change to multiple variables at the solution representation. Here, we first show that DO has impressive performance compared with other leading MBOAs (and other rival methods) on multiple knapsack problems, a standard combinatorial optimization problem of general interest. Going deeper, we then carry out a detailed investigation to understand the differences between DO and other MBOAs, identifying key problem characteristics where other MBOAs are afflicted by exponential running times, and DO is not. This study serves to concretize our understanding of the Major Transitions, and why that leads to evolvability, and also provides a strong motivation for further investigation of deep learning methods in optimization.

ABSTRACT. This paper focuses on adapting and applying a genetic algorithm (GA) and differential evolution (DE) to solve the grain (wheat) mixing problem. The proposed algorithms explore a search space that aims at finding a quality mixing of wheat from grain bins that produce the maximum profit at a grain elevator. The experimental results demonstrate that mixing bins provide more profit than not mixing, and that the evolutionary approaches lead to consistently higher profits than the non-evolutionary methods.

ABSTRACT. Management and analyses of water resources is of paramount importance in the implementation of water related sustainable development goals. Hydraulic models are key in flood forecasting and simulation applied to a river flood analysis and risk prediction and an accurate estimation of the roughness is one of the main factors in predicting the discharge in a stream. In practical implementation roughness can be represented by the prediction of the well known Manning's coefficient necessary for discharge calculation. In this paper we design an objective function that measures the quality of a given configuration of the Manning's coefficient. Such an objective function is optimised through several evolutionary approaches, namely: (1+1)-ES, CMA-ES, Differential Evolution, Particle Swarm Optimization and Bayesian Optimization. As case of study, a river in the central Italy was considered. The results indicate that the model, consistent with the classical techniques adopted in the hydraulic engineering field, is applicable to natural rivers and is able to provide an estimation of the roughness coefficients with a satisfactory accuracy. A comparison of the performances of the five evolutionary algorithms is also proposed.

ABSTRACT. The problem of pollution in water nowadays, is a very serious issue to be faced and it is really important to be able to monitor it with non-invasive and low-cost solutions, like those offered by smart sensor technologies. In this paper, we propose an improvement of our innovative classification system, based on geometrical cones, to detect and classify pollutants, belonging to a given set of substances, spilled into wastewater. The solution is based on an ad-hoc classifier that can be implemented abord the Smart Cable Water (SCW) sensor, based on SENSIPLUS technology developed by Sensichips s.r.l. The SCW is a smart-sensor endowed with six interdigitated electrodes, covered by specific sensing materials that allow detecting between different water contaminants. In order to develop an algorithm suitable to apply the "edge computing" paradigm we first compress the input data from a 10-dimensional space to a 3-D space by using the PCA decomposition techniques. Then we use an ad-hoc classifier to classify between the different contaminants in the transformed space. To learn the classifier's parameters we used the evolutionary algorithms. The obtained results have been compared with the old classification system and other, more classical, machine and deep learning approaches.

ABSTRACT. This paper proposes the duplex route generation method to evolve the bus route network which is robust to environmental changes and aims at investigating its effectiveness through the experiments. In this study, the “duplex route” corresponds to the alternative route and it has an advantage of not requiring to modify the route network in the environmental changes. To generate the duplex routes, this study employs MOEA/D as the base optimization method and introduces the following two operations in MOEA/D to increase the duplex routes while improving the fitness: (1) the crossover operation to generate the duplex routes, which is improved from the crossover operation in SEAMO2 [10] that evolves unique routes, and (2) the priority solution update operation in the enhanced MOEA/D [5] to maintain diversity of the routes which contributes to improving the fitness. The experiments on the Mandl’s benchmark problem has revealed: (1) the proposed crossover operation can generate many duplex networks as compared to the original crossover operation; (2) the priority solution update operation improves the fitness, i.e., a minimization of the passenger transportation time and the number of buses; and (3) an integration of the two operations improves both the number of duplex routes and fitness, which is hard to be achieved by either operation.