Spiking Neural Networks (SNNs) have attracted recent interest due to their energy efficiency and ... more Spiking Neural Networks (SNNs) have attracted recent interest due to their energy efficiency and biological plausibility. However, the performance of SNNs still lags behind traditional Artificial Neural Networks (ANNs), as there is no consensus on the best learning algorithm for SNNs. Best-performing SNNs are based on ANN to SNN conversion or learning with spike-based backpropagation through surrogate gradients. The focus of recent research has been on developing and testing different learning strategies, with hand-tailored architectures and parameter tuning. Neuroevolution (NE), has proven successful as a way to automatically design ANNs and tune parameters, but its applications to SNNs are still at an early stage. DENSER is a NE framework for the automatic design and parametrization of ANNs, based on the principles of Genetic Algorithms (GA) and Structured Grammatical Evolution (SGE). In this paper, we propose SPENSER, a NE framework for SNN generation based on DENSER, for image classification on the MNIST and Fashion-MNIST datasets. SPENSER generates competitive performing networks with a test accuracy of 99.42% and 91.65% respectively. • Computing methodologies → Neural networks; Computer vision; • Theory of computation → Evolutionary algorithms; Grammars and context-free languages.
Typefaces are an essential resource employed by graphic designers. The increasing demand for inno... more Typefaces are an essential resource employed by graphic designers. The increasing demand for innovative type design work increases the need for good technological means to assist the designer in the creation of a typeface. We present an evolutionary approach for the automatic generation of type stencils to draw coherent glyphs for different characters. The proposed system employs a Genetic Algorithm to evolve populations of type stencils. The evaluation of each candidate stencil uses a hill climbing algorithm to search the best configurations to draw the target glyphs. We study the interplay between legibility, coherence and expressiveness, and show how our framework can be used in practice.
Proceedings of the Genetic and Evolutionary Computation Conference
Locating odour sources with mobile robots is a difficult task with many applications. Over the ye... more Locating odour sources with mobile robots is a difficult task with many applications. Over the years, researchers have devised bioinspired and cognitive methods to enable mobile robots to fulfil this task. Cognitive approaches are effective in large spaces, but computationally heavy. On the other hand, bio-inspired ones are lightweight, but they are only effective in the presence of frequent stimuli. One of the most popular cognitive approaches is Infotaxis, which iteratively computes a probability map of the source location. Another strand of work uses Genetic Programming to produce complete search strategies from bio-inspired behaviours. This work combines the two approaches by allowing Genetic Programming to evolve search strategies that include infotactic and bio-inspired behaviours. The proposed method is tested in a set of environments with distinct airflow and chemical dispersion patterns. Its performance is compared to that of evolved strategies without infotactic behaviours and to the standard infotaxis approach. The statistically validated results show that the proposed method produces search strategies that have significantly higher success rates, whilst being faster than those produced by any of the original approaches. Moreover, the best evolved strategies are analysed, providing insight into when infotaxis is more beneficial. • Computing methodologies → Artificial intelligence; • Theory of computation → Design and analysis of algorithms; • Computer systems organization → Robotics.
Practical Applications of Computational Biology and Bioinformatics, 12th International Conference, 2018
Genetic programming represents a flexible and powerful evolutionary technique in machine learning... more Genetic programming represents a flexible and powerful evolutionary technique in machine learning. The use of genetic programming for rule induction has generated interesting results in classification problems. This paper proposes an evolutionary approach for logical rule induction, which is applied to clinical data. Since logical rules disclose knowledge from the analyzed data, we use such a knowledge to filter features from the target dataset. The results reached by the used dataset have been very promising when used in classification tasks and compared with other methods.
Evolutionary and Biologically Inspired Music, Sound, Art and Design, 2016
Over the years researchers have been interested in devising computational approaches for music an... more Over the years researchers have been interested in devising computational approaches for music and image generation. Some of the approaches rely on generative rewriting systems like L-systems. More recently, some authors questioned the interplay of music and images, that is, how we can use one type to drive the other. In this paper we present a new method for the algorithmic generations of images that are the result of a visual interpretation of an L-system. The main novelty of our approach is based on the fact that the L-system itself is the result of an evolutionary process guided by musical elements. Musical notes are decomposed into elements -pitch, duration and volume in the current implementation -and each of them is mapped into corresponding parameters of the L-system -currently line length, width, color and turning angle. We describe the architecture of our system, based on a multi-agent simulation environment, and show the results of some experiments that provide support to our approach.
Evolutionary Algorithms (EA) are a family of search heuristics from the area of Artificial Intell... more Evolutionary Algorithms (EA) are a family of search heuristics from the area of Artificial Intelligence. They have been successfully applied in problems of learning, optimization and design, from many application domains. Currently, they are divided into two families, Genetic Algorithms (GA) and Genetic Programming (GP). Genetic Algorithms evolve solutions for a specific problem. On the other hand, Genetic Programming evolves programs that, when executed, produce the solutions for specific problems. Many of the successful applications of EAs have been on static environments, i.e., environments whose conditions remain constant throughout time. However, many real world applications involve dynamic environments, meaning that the problems themselves change over time. The difficulty of evolving solutions in dynamic environments emerges from a common problem of EAs known as premature convergence. This phenomenon happens when the population converges to a good quality area of the search space, being the individuals very similar to each other. In static environments, this may cause the algorithm to only find local optima instead of the global optimum solution. On the other hand, in dynamic environments, this phenomenon may cause a greater difficulty and delay in finding good solutions when the environment changes, specially if the new environment is very different from the previous one. There is already some work on adapting GAs for evolving solutions in dynamic environments. However, the same can not be said for Genetic Programming. The goal of this thesis is to fill that gap. We will do so by transposing some of the existing mechanisms for GAs to GPs. Moreover, we will propose novel approaches, that have not yet been employed in GPs. We will test the developed algorithms in three well known benchmark problems, with different types of dynamic environments, and proceed to do a statistical analysis of the collected data.
An evolutionary algorithm designed to successfully search for Optimal Golomb Rulers is presented.... more An evolutionary algorithm designed to successfully search for Optimal Golomb Rulers is presented. The proposed approach uses a binary representation to codify the marks contained in a ruler. Standard genetic operators are used. During evaluation, insertion and correction procedures are applied in order to improve the algorithm performance. Experimental results show that this approach is effective and capable of identifying good solutions. Furthermore, a comprehensive study is performed to understand the role of insertion and correction. Results reveal that the first method is essential to the success of the search process, whereas the importance of the second one remains unclear.
Proceedings of the 2000 Congress on Evolutionary Computation. CEC00 (Cat. No.00TH8512), 2000
The wide use of fieldbus based distributed systems in embedded control applications triggered the... more The wide use of fieldbus based distributed systems in embedded control applications triggered the research on the problem of transmission network induced jitter in control variables. In this paper we introduce a variant of the classical Genetic Algorithm, which we call Progressive Genetic Algorithm, and show how it can be used to reduce jitter suffered by periodic messages. The approach can be applied either in centrally controlled fieldbuses or in synchronized ones. The algorithm was tested with two well-known and widely used benchmarks: the PSA, coming from automotive industries and the SAE from Automatic Guided Vehicles. It is shown that it is possible to eliminate completely jitter if the adequate transmission rate is available and, if not, a satisfactory reduced jitter can be obtained.
This study introduces the particle swarm metaphor to the domain of organizational adaptation. A s... more This study introduces the particle swarm metaphor to the domain of organizational adaptation. A simulation model (OrgSwarm) is constructed to examine the impact of strategic inertia, in the presence of errorful assessments of future payoffs to potential strategies, on the adaptation of the strategic fitness of a population of organizations. The results indicate that agent (organization) uncertainty as to the payoffs of potential strategies has the affect of lowering average payoffs obtained by a population of organizations. The results also indicate that a degree of strategic inertia, in the presence of an election mechanism, assists rather than hampers adaptive efforts in static and slowly changing strategic environments.
Creativity is one of the most remarkable characteristics of the human mind. It is thus natural th... more Creativity is one of the most remarkable characteristics of the human mind. It is thus natural that Artificial Intelligence's research groups have been working towards the study and proposal of adequate computational models to creativity. Artificial creative systems are potentially effective in a wide range of artistic, architectural and engineering domains where detailed problem specification is virtually impossible and, therefore, conventional problem solving is unlikely to produce useful solutions. Moreover their study may contribute to the overall understanding of the mechanisms behind human creativity. In this text, we propose a computational hybrid architecture for creative reasoning aimed at empowering cross contributions from Case-Based Reasoning (CBR) and Evolutionary Computation (EC). The first will provide us a long-term memory, while the later will complement with its adaptive ability. The background knowledge provided by the memory mechanism can be exploited to solve problems inside the same domain or problems that imply inter-domain transfer of expertise. The architecture is the result of a synthesis work motivated by the observation that the strong similarities between the computational mechanisms used in systems developed so far could be explored. Moreover, we also propose that those mechanisms may be supported by a common knowledge representation formalism, which appears to be adequate to a considerable range of domains. Furthermore, we consider that this architecture may be explored as a unifying model for the creative process, contributing to the deepening of the theoretical foundations of the area.
2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational Intelligence), 2008
Fitness landscape analysis techniques are used to better understand the influence of genetic repr... more Fitness landscape analysis techniques are used to better understand the influence of genetic representations and associated variation operators when solving a combinatorial optimization problem. Several representations for the Optimal Golomb Ruler problem are examined. Common mutation operators such as bit-flip mutation are employed to generate fitness landscapes to study the genetic representations. Furthermore, additional experiments are made to observe the effects of adding heuristics and local improvements to the encodings.
Proceedings of the 7th annual conference on Genetic and evolutionary computation, 2005
Resource-Limited Genetic Programming is a bloat control technique that imposes a single limit on ... more Resource-Limited Genetic Programming is a bloat control technique that imposes a single limit on the total amount of resources available to the entire population, where resources are tree nodes or code lines. We elaborate on this recent concept, introducing a dynamic approach to managing the amount of resources available for each generation. Initially low, this amount is increased only if it results in better population fitness. We compare the dynamic approach to the static method where a constant amount of resources is available throughout the run, and with the most traditional usage of a depth limit at the individual level. The dynamic approach does not impair performance on the Symbolic Regression of the quartic polynomial, and achieves excellent results on the Santa Fe Artificial Ant problem, obtaining the same fitness with only a small percentage of the computational effort demanded by the other techniques.
Proceedings of the 2003 ACM symposium on Applied computing - SAC '03, 2003
A comparative study is made between a new evolutionary approach for the Vehicle Routing Problem (... more A comparative study is made between a new evolutionary approach for the Vehicle Routing Problem (VRP) and a standard evolutionary model, based on Path Representation. Genetic Vehicle Representation (GVR) is the new twolevel representational scheme designed to deal in an effective way with all the information needed by candidate solutions. Experimental results, obtained from a set of VRP instances, show performance improvements when GVR is used.
In this paper we analyse a new evolutionary approach to the vehicle routing problem. We present G... more In this paper we analyse a new evolutionary approach to the vehicle routing problem. We present Genetic Vehicle Representation (GVR), a two-level representational scheme designed to deal in an effective way with all the information that candidate solutions must encode. Experimental results show that this method is both effective and robust, allowing the discovery of new best solutions for some well-known benchmarks.
We propose replacing the traditional tree depth limit in Genetic Programming by a single limit on... more We propose replacing the traditional tree depth limit in Genetic Programming by a single limit on the amount of resources available to the whole population, where resources are the tree nodes. The resource-limited technique removes the disadvantages of using depth limits at the individual level, while introducing automatic population resizing, a natural side-effect of using an approach at the population level. The results show that the replacement of individual depth limits by a population resource limit can be done without impairing performance, thus validating this first and important step towards a new approach to improving the efficiency of GP.
Transformational creativity requires a change of the search space. As such, Evolutionary Computat... more Transformational creativity requires a change of the search space. As such, Evolutionary Computation (EC) approaches are incapable of transformational creativity. In this paper, we discuss how canonical EC techniques can be extended in order to yield the potential of transformational creativity. We present a formalized description of how this can be attained, and the experimental results achieved with a metaevolutionary scheme.
In this paper we present a new evolutionary algorithm designed to efficiently search for optimal ... more In this paper we present a new evolutionary algorithm designed to efficiently search for optimal Golomb rulers. The proposed approach uses a redundant random keys representation to codify the information contained in a chromosome and relies on a simple interpretation algorithm to obtain feasible solutions. Experimental results show that this method is successful in quickly identifying good solutions and that can be considered as a realistic alternative to massive parallel approaches that need several months or years to discover high quality Golomb rulers.
When using Evolutionary Algorithms (EAs) in no stationary problems some extensions have been intr... more When using Evolutionary Algorithms (EAs) in no stationary problems some extensions have been introduced in order to avoid the convergence of the population towards a point of the search space. One of these improvements consists in the use of explicit memory responsible for storing good individuals from the search population. When the environment is cyclic and previous environments reappear later, memory should allow continuous progression of the EA's performance with the least decline of the individuals' fitness. But in most situations this purpose is not achieved, and the typical behavior of an EA when a change happens is the bestfitness decrease and some time is necessary to readapting to the new conditions. The key problem when using explicit memory is the size's restrictions usually imposed. So, when it is necessary to store a new individual and memory is full, it is necessary to replace individuals. This replacement can lead to the destruction of information that might be useful in the future. In this work we are interested in the enhancement of memory's usage and we propose two new replacing methods to apply when memory is full. The investigated methods are tested in several memory-based EAs and the obtained results show that memory can be used in a more effective way such that the algorithms' performance is strongly improved.
The goal of this research is to analyze how individual learning interacts with an evolutionary al... more The goal of this research is to analyze how individual learning interacts with an evolutionary algorithm in its search for best candidates for the Busy Beaver problem. To study this interaction two learning models, implemented as local search procedures, are proposed. Experimental results show that, in highly irregular and prone to premature convergence search spaces, local search methods are not an effective help to evolution. In addition, one interesting effect related to learning is reported. When the mutation rate is too high, learning acts as a repair, reintroducing some useful information that was lost.
Uploads
Papers by Ernesto Costa