Search

Feature selection is the process of choosing, or removing, features to obtain the most informative feature subset of minimal size. Such subsets are used to improve performance of machine learning algorithms and enable human understanding of the results. Approaches to feature selection in literature exploit several optimization algorithms. Multi-objective methods also have been proposed, minimizing at the same time the number of features and the error. While most approaches assess error resorting to the average of a stochastic K-fold cross-validation, comparing averages might be misleading. In this paper, we show how feature subsets with different average error might in fact be non-separable when compared using a statistical test. Following this idea, clusters of non-separable optimal feature subsets are identified. The performance in feature selection can thus be evaluated by verifying how many of these optimal feature subsets an algorithm is able to identify. We thus propose a multi-objective optimization approach to feature selection, EvoFS, with the objectives to i. minimize feature subset size, ii. minimize test error on a 10-fold cross-validation using a specific classifier, iii. maximize the analysis of variance value of the lowest-performing feature in the set. Experiments on classification datasets whose feature subsets can be exhaustively evaluated show that our approach is able to always find the best feature subsets. Further experiments on a high-dimensional classification dataset, that cannot be exhaustively analyzed, show that our approach is able to find more optimal feature subsets than state-of-the-art feature selection algorithms.

In the field of artificial intelligence, agents learn how to take decisions by fitting their parameters on a set of samples called training set. Similarly, a core set is a subset of the training samples such that, if an agent exploits this set to fit its parameters instead of the whole training set, then the quality of the inferences does not change significantly. Relaxing the constraint that restricts the search for core sets to the available data, neural networks may be used to generate virtual samples, called archetype set, containing the same kind of information. This work illustrates the features of GH-ARCH, a recently proposed self-organizing hierarchical neural network for archetype discovery. Experiments show how the use of archetypes allows both ML agents to make fast and accurate predictions and human experts to make sense of such decisions by analyzing few important samples.

Digital collectible card games are not only a growing part of the video game industry, but also an interesting research area for the field of computational intelligence. This game genre allows researchers to deal with hidden information, uncertainty and planning, among other aspects. This paper proposes the use of evolutionary algorithms (EAs) to develop agents who play a card game, Hearthstone, by optimizing a data-driven decision-making mechanism that takes into account all the elements currently in play. Agents feature self-learning by means of a competitive coevolutionary training approach, whereby no external sparring element defined by the user is required for the optimization process. One of the agents developed through the proposed approach was runner-up (best 6%) in an international Hearthstone Artificial Intelligence (AI) competition. Our proposal performed remarkably well, even when it faced state-of-the-art techniques that attempted to take into account future game states, such as Monte-Carlo Tree search. This outcome shows how evolutionary computation could represent a considerable advantage in developing AIs for collectible card games such as Hearthstone.

Machine learning agents learn to take decisions extracting information from training data. When similar inferences can be obtained using a small subset of the same training set of samples, the subset is called coreset. Coresets discovery is an active line of research as it may be used to reduce the training speed as well as to allow human experts to gain a better understanding of both the phenomenon and the decisions, by reducing the number of samples to be examined. For classification problems, the state-of-the-art in coreset discovery is EvoCore, a multi-objective evolutionary algorithm. In this work EvoCore is exploited both on synthetic and on real data sets, showing how coresets may be useful in explaining decisions taken by machine learning classifiers.

Industrial manipulators are robots used to replace humans in dangerous or repetitive tasks. Also, these devices are often used for applications where high precision and accuracy is required. The increase of backlash caused by wear, that is, the increase of the amount by which teeth space exceeds the thickness of gear teeth, might be a significant problem, that could lead to impaired performances or even abrupt failures. However, maintenance is difficult to schedule because backlash cannot be directly measured and its effects only appear in closed loops. This paper proposes a novel technique, based on an Evolutionary Algorithm, to estimate the increase of backlash in a robot joint transmission. The peculiarity of this method is that it only requires measurements from the motor encoder. Experimental evaluation on a real-world test case demonstrates the effectiveness of the approach.

In the field of machine learning, coresets are defined as subsets of the training set that can be used to obtain a good approximation of the behavior that a given algorithm would have on the whole training set. Advantages of using coresets instead of the training set include improving training speed and allowing for a better human understanding of the dataset. Not surprisingly, coreset discovery is an active research line, with several notable contributions in literature. Nevertheless, restricting the search for representative samples to the available data points might impair the final result. In this work, neural networks are used to create sets of virtual data points, named archetypes, with the objective to represent the information contained in a training set, in the same way a coreset does. Starting from a given training set, a hierarchical clustering neural network is trained and the weight vectors of the leaves are used as archetypes on which the classifiers are trained. Experimental results on several benchmarks show that the proposed approach is competitive with traditional coreset discovery techniques, delivering results with higher accuracy, and showing a greater ability to generalize to unseen test data.

Social networks are one the main sources of information transmission nowadays. However, not all nodes in social networks are equal: in fact, some nodes are more influential than others, i.e., their information tends to spread more. Finding the most influential nodes in a network – the so-called Influence Maximization problem – is an NP-hard problem with great social and economical implications. Here, we introduce a framework based on Evolutionary Algorithms that includes various graph-aware techniques (spread approximations, domain-specific operators, and node filtering) that facilitate the optimization process. The framework can be applied straightforwardly to various social network datasets, e.g., those in the SNAP repository.

International audience

This paper gives an overview of scientific challenges that occur when performing life-cycle assessment (LCA) in the food supply chain. In order to evaluate these risks, the Failure Mode and Effect Analysis tool has been used. Challenges related to setting the goal and scope of LCA revealed four hot spots: system boundaries of LCA; used functional units; type and quality of data categories, and main assumptions and limitations of the study. Within the inventory analysis, challenging issues are associated with allocation of material and energy flows and waste streams released to the environment. Impact assessment brings uncertainties in choosing appropriate environmental impacts. Finally, in order to interpret results, a scientifically sound sensitivity analysis should be performed to check how stable calculations and results are. Identified challenges pave the way for improving LCA of food supply chains in order to enable comparison of results.

In machine learning a coreset is defined as a subset of the training set using which an algorithm obtains performances similar to what it would deliver if trained over the whole original data. Advantages of coresets include improving training speed and easing human understanding. Coreset discovery is an open line of research as limiting the training might also impair the quality of the result. Differently, virtual points, here called archetypes, might be far more informative for a machine learning algorithm. Starting from this intuition, a novel evolutionary approach to archetype set discovery is presented: starting from a population seeded with candidate coresets, a multi-objective evolutionary algorithm is set to modify them and eventually create archetype sets, to minimize both number of points in the set and classification error. Experimental results on popular benchmarks show that the proposed approach is able to deliver results that allow a classifier to obtain lower error and better ability of generalizing on unseen data than state-of-the-art coreset discovery techniques.

In an optimization problem, a coreset can be defined as a subset of the input points, such that a good approximation to the optimization problem can be obtained by solving it directly on the coreset, instead of using the whole original input. In machine learning, coresets are exploited for applications ranging from speeding up training time, to helping humans understand the fundamental properties of a class, by considering only a few meaningful samples. The problem of discovering coresets, starting from a dataset and an application, can be defined as identifying the minimal amount of samples that do not significantly lower performance with respect to the performance on the whole dataset. Specialized literature offers several approaches to finding coresets, but such algorithms often disregard the application, or explicitly ask the user for the desired number of points. Starting from the consideration that finding coresets is an intuitively multi-objective problem, as minimizing the number of points goes against maintaining the original performance, in this paper we propose a multi-objective evolutionary approach to identifying coresets for classification. The proposed approach is tested on classical machine learning classification benchmarks, using 6 state-of-the-art classifiers, comparing against 7 algorithms for coreset discovery. Results show that not only the proposed approach is able to find coresets representing different compromises between compactness and performance, but that different coresets are identified for different classifiers, reinforcing the assumption that coresets might be closely linked to the specific application.

Background MicroRNAs (miRNAs) are noncoding RNA molecules heavily involved in human tumors, in which few of them circulating the human body. Finding a tumor-associated signature of miRNA, that is, the minimum miRNA entities to be measured for discriminating both different types of cancer and normal tissues, is of utmost importance. Feature selection techniques applied in machine learning can help however they often provide naive or biased results. Results An ensemble feature selection strategy for miRNA signatures is proposed. miRNAs are chosen based on consensus on feature relevance from high-accuracy classifiers of different typologies. This methodology aims to identify signatures that are considerably more robust and reliable when used in clinically relevant prediction tasks. Using the proposed method, a 100-miRNA signature is identified in a dataset of 8023 samples, extracted from TCGA. When running eight-state-of-the-art classifiers along with the 100-miRNA signature against the original 1046 features, it could be detected that global accuracy differs only by 1.4%. Importantly, this 100-miRNA signature is sufficient to distinguish between tumor and normal tissues. The approach is then compared against other feature selection methods, such as UFS, RFE, EN, LASSO, Genetic Algorithms, and EFS-CLA. The proposed approach provides better accuracy when tested on a 10-fold cross-validation with different classifiers and it is applied to several GEO datasets across different platforms with some classifiers showing more than 90% classification accuracy, which proves its cross-platform applicability. Conclusions The 100-miRNA signature is sufficiently stable to provide almost the same classification accuracy as the complete TCGA dataset, and it is further validated on several GEO datasets, across different types of cancer and platforms. Furthermore, a bibliographic analysis confirms that 77 out of the 100 miRNAs in the signature appear in lists of circulating miRNAs used in cancer studies, in stem-loop or mature-sequence form. The remaining 23 miRNAs offer potentially promising avenues for future research.

When a machine learning algorithm is able to obtain the same performance given a complete training set, and a small subset of samples from the same training set, the subset is termed coreset. As using a coreset improves training speed and allows human experts to gain a better understanding of the data, by reducing the number of samples to be examined, coreset discovery is an active line of research. Often in literature the problem of coreset discovery is framed as i. single-objective, attempting to find the candidate coreset that best represents the training set, and ii. independent from the machine learning algorithm used. In this work, an approach to evolutionary coreset discovery is presented. Building on preliminary results, the proposed approach uses a multi-objective evolutionary algorithm to find compromises between two conflicting objectives, i. minimizing the number of samples in a candidate coreset, and ii. maximizing the accuracy of a target classifier, trained with the coreset, on the whole original training set. Experimental results on popular classification benchmarks show that the proposed approach is able to identify candidate coresets with better accuracy and generality than state-of-the-art coreset discovery algorithms found in literature.

International audience; Background: Agri-food is one of the most important sectors of the industry in Europe and potentially a major contributor to the global warming. Sustainability issues in this context pose a huge challenge for several reasons: the variety of considered scales, the number of disciplines involved, the uncertainties, the out-of-equilibrium states, the complex quantitative and qualitative factors, the normative issues and the availability of data. Although important insight and breakthroughs have been attained in different scientific domains, an overarching and integrated analysis of these complex problems have yet to be realized.Scope and Approach: This context creates huge opportunities for research in interaction with mathematical programming, integrative models and decision-support tools. The paper propose a computational viewpoint including questions of holistic approach, multiscale reconstruction and optimization. Some directions are discussed.Key Findings and Conclusions: Several research questions based on a mathematical programming framework are emerging: how can such a framework manage uncertainty, cope with complex qualitative and quantitative information essential for social and environmental considerations, encompass diverse scales in space and time, cope with a multivariable dynamic environment and with scarcity of data. Moreover, how can it deal with different perspectives, types of models, research goals and data produced by conceptually disjoint scientific disciplines, ranging from physics and physiology to sociology and ethics? Building models is essential, but highly difficult; it will need a strong iterative interaction combining computational intensive methods, formal reasoning and the experts of the different fields. Some future research directions are proposed, involving all those dimensions: mathematical resilience, human-machine interactive learning and optimization techniques. (C) 2015 Elsevier Ltd. All rights reserved.

International audience; VALIS is an effective and robust classification algorithm with a focus on understandability. Its name stems from Vote-ALlocating Immune System, as it evolves a population of artificial antibodies that can bind to the input data, and performs classification through a voting process. In the beginning of the training, VALIS generates a set of random candidate antibodies; at each iteration, it selects the most useful ones to produce new candidates, while the least, are discarded; the process is iterated until a user-defined stopping condition. The paradigm allows the user to get a visual insight of the learning dynamics, helping to supervise the process, pinpoint problems, and tweak feature engineering. VALIS is tested against nine state-of-the-art classification algorithms on six popular benchmark problems; results demonstrate that it is competitive with well-established black-box techniques, and superior in specific corner cases.

A better understanding of protein fouling during the thermal treatment of whey protein concentrate (WPC) solutions is critical for better fouling control. In order to understand the impact of various parameters on the total whey protein fouling mass, a dimensional analysis was applied to the experimental data obtained from a pilot scale plate heat exchanger, setting total fouling mass as the target variable. A model was developed to predict the total fouling mass, covering a series of variables including whey protein solution concentration (2.5–25 g/L), calcium concentration (70-120 ppm), running time (90-330 min), fouling solution flow rate (200-500 L/h), total fouling surface area, outlet temperature (82-97 °C) and differences in whey protein concentrate powders. In addition to temperature dimensionless parameters, the main parameters involved in the model are the Reynolds number (2000-5000) and the calcium to β-lactoglobulin molar ratio (2.7–34.7). The model developed concerns only pure whey proteins solutions since all the testing solutions were casein free. This model has allowed us to provide guidelines as to how the above parameters influence fouling within the plate heat exchanger, as well as empirical correlations for predicting such fouling development.

International audience; Collectible card games have been among the most popular and profitable products of the entertainment industry since the early days of Magic: The GatheringTM in the nineties. Digital versions have also appeared, with HearthStone: Heroes of WarCraftTM being one of the most popular. In Hearthstone, every player can play as a hero, from a set of nine, and build his/her deck before the game from a big pool of available cards, including both neutral and hero-specific cards. This kind of games offers several challenges for researchers in artificial intelligence since they involve hidden information, unpredictable behaviour, and a large and rugged search space. Besides, an important part of player engagement in such games is a periodical input of new cards in the system, which mainly opens the door to new strategies for the players. Playtesting is the method used to check the new card sets for possible design flaws, and it is usually performed manually or via exhaustive search; in the case of Hearthstone, such test plays must take into account the chosen hero, with its specific kind of cards. In this paper, we present a novel idea to improve and accelerate the playtesting process, systematically exploring the space of possible decks using an Evolutionary Algorithm (EA). This EA creates HearthStone decks which are then played by an AI versus established human-designed decks. Since the space of possible combinations that are play-tested is huge, search through the space of possible decks has been shortened via a new heuristic mutation operator, which is based on the behaviour of human players modifying their decks. Results show the viability of our method for exploring the space of possible decks and automating the play-testing phase of game design. The resulting decks, that have been examined for balancedness by an expert player, outperform human-made ones when played by the AI; the introduction of the new heuristic operator helps to improve the obtained solutions, and basing the study on the whole set of heroes shows its validity through the whole range of decks.

One of the most relevant problems in social networks is influence maximization, that is the problem of finding the set of the most influential nodes in a network, for a given influence propagation model. As the problem is NP-hard, recent works have attempted to solve it by means of computational intelligence approaches, for instance Evolutionary Algorithms. However, most of these methods are of limited applicability for real-world large-scale networks, for two reasons: on the one hand, they require a large number of candidate solution evaluations to converge; on the other hand, each evaluation is computationally expensive in that it needs a considerable number of Monte Carlo simulations to obtain reliable values. In this work, we consider a possible solution to such limitations, by evaluating a surrogate-assisted Multi-Objective Evolutionary Algorithm that uses an approximate model of influence propagation (instead of Monte Carlo simulations) to find the minimum-sized set of most influential nodes. Experiments carried out on two social networks datasets suggest that approximate models should be carefully considered before using them in influence maximization approaches, as the errors induced by these models are in some cases too big to benefit the algorithmic performance.

The production system of freeze-dried lactic acid bacteria involves several processes, but its impact on bacteria resistance is still not well understood. This system can be defined as a complex one since it depends on multiple scales: the Genomic, the Cellular and the Population scale. The scarcity of data available for building models leads us to propose an approach that makes use of expert knowledge. In this paper we present a semiautomatic modelling tool, LIDEOGRAM and discuss how it contributes to insight formulation and rapid hypothesis testing. New results show that LIDEOGRAM is able to produce more robust modelling hypotheses when experts can interact and revisit the genomic data preprocessing.

International audience; Cancer diagnosis is currently undergoing a paradigm shift with the incorporation of molecular biomarkers as part of routine diagnostic panel. This breakthrough discovery directs researches to examine the role of microRNA in cancer, since its deregulation is often associated with almost all human tumors. Such differences frequently recur in tumor-specific microRNA signatures, which are helpful to diagnose tissue of origin and tumor subtypes. Nonetheless, the resulting classification problem is far from trivial, as there are hundreds of microRNA types, and tumors are non-linearly correlated to the presence of several overexpressions. In this paper, we propose to apply an evolutionary optimized convolutional neural network classifier to this complex task. The presented approach is compared against 21 state-of-the-art classifiers, on a real-world dataset featuring 8129 patients, for 29 different classes of tumors, using 1046 different biomarkers. As a result of the comparison, we also present a meta-analysis on the dataset, identifying the classes on which the collective performance of the considered classifiers is less effective, and thus possibly singling out types of tumors for which biomarker tests might be less reliable.

Since centuries, agriculture, food and biological systems are strongly linked to human expertise, albeit such knowledge has been capitalized and shared often at a local level, only. Since the beginning of the last century, swept away by productivism, modern agriculture and food production have put cumulated human knowledge aside. Facing new challenges like sustainability in a changing context, holistic approaches cannot be managed “manually” ab initio and there is a clear need for computing decision-support tools to tackle these new issues. Moreover, new approaches should be built centred on humans and for humans. The heart of our purpose is to shift the focus again on human and local expertise, guided by powerful computing interactive systems.

This article provides an overview of the 6 workshops held in conjunction with PPSN 2018 in Coimbra, Portugal. For each workshop, we list title, organizers, aim and scope as well as the accepted contributions.

International audience; Reducing the effort required by humans in countering malware is of utmost practical value. We describe a scalable, semi-supervised framework to dig into massive data sets of Android applications and identify new malware families. Until 2010, the industrial standard for the detection of malicious applications has been mainly based on signatures; as each tiny alteration in malware makes them ineffective, new signatures are frequently created - a task that requires a considerable amount of time and resources from skilled experts. The framework we propose is able to automatically cluster applications in families and suggest formal rules for identifying them with 100% recall and quite high precision. The families are used either to safely extend experts' knowledge on new samples or to reduce the number of applications requiring thorough analyses. We demonstrated the effectiveness and the scalability of the approach running experiments on a database of 1.5 million Android applications. In 2018, the framework has been successfully deployed on Koodous, a collaborative anti-malware platform.

International audience; This paper presents a novel model of protein hydrolysis and release of peptides by endoproteases. It requires the amino-acid sequence of the protein substrate to run, and makes use of simple Monte-Carlo in silico simulations to qualitatively and quantitatively predict the peptides that are likely to be produced during the course of the proteolytic reaction. In the present study, the model is applied to the case of pepsin, the gastric protease. Unlike pancreatic proteases, pepsin has a low substrate specificity and therefore displays a stochastic behavior that is particularly challenging to model and predict. Two versions of the model are studied and compared with peptidomic data obtained during pepsin hydrolysis of bovine lactoferrin. The first version of the model takes into account cleavage probabilities according to the amino acids in position P1-P1' only, whereas the second version also accounts for the influence of neighbor amino acids (P4, P3, P2, P2', P3', P4') and peptide terminal ends. The second version of the model was able to reproduce many real-world features of the reported behavior of pepsin, such as the peptide size distribution, or the quantity of free amino-acids. More remarkably, 50% of the experimentally monitored peptides (44/87) lay within the 120 most abundant simulated peptides. The presented methodology has the advantage of being applicable not only to different proteins, but to different enzymes as well, as long as cleavage frequency data are available.

In this paper the application of machine learning techniques for the development of constitutive material models is being investigated. A flow stress model, for strain rates ranging from 10 −4 to 1012 (quasi-static to highly dynamic), and temperatures ranging from room temperature to over 1000 K, is obtained by beginning directly with experimental stress–strain data for Copper. An incrementally objective and fully implicit time integration scheme is employed to integrate the hypo-elastic constitutive model, which is then implemented into a finite element code for evaluation. Accuracy and performance of the flow stress models derived from symbolic regression are assessed by comparison to Taylor anvil impact data. The results obtained with the free-form constitutive material model are compared to well-established strength models such as the Preston–Tonks–Wallace (PTW) model and the Mechanical Threshold Stress (MTS) model. Preliminary results show candidate free-form models comparing well with data in regions of stress–strain space with sufficient experimental data, pointing to a potential means for both rapid prototyping in future model development, as well as the use of machine learning in capturing more data as a guide for more advanced model development.

The two volumes LNCS 10199 and 10200 constitute the refereed conference proceedings of the 20th European Conference on the Applications of Evolutionary Computation, EvoApplications 2017, held in Amsterdam, The Netherlands, in April 2017, collocated with the Evo* 2016 events EuroGP, EvoCOP, and EvoMUSART. The 46 revised full papers presented together with 26 poster papers were carefully reviewed and selected from 108 submissions. EvoApplications 2016 consisted of the following 13 tracks: EvoBAFIN (natural computing methods in business analytics and finance), EvoBIO (evolutionary computation, machine learning and data mining in computational biology), EvoCOMNET (nature-inspired techniques for telecommunication networks and other parallel and distributed systems), EvoCOMPLEX (evolutionary algorithms and complex systems), EvoENERGY (evolutionary computation in energy applications), EvoGAMES (bio-inspired algorithms in games), EvoIASP (evolutionary computation in image analysis, signal processing, and pattern recognition), EvoINDUSTRY (nature-inspired techniques in industrial settings), EvoKNOW (knowledge incorporation in evolutionary computation), EvoNUM (bio-inspired algorithms for continuous parameter optimization), EvoPAR (parallel implementation of evolutionary algorithms), EvoROBOT (evolutionary robotics), EvoSET (nature-inspired algorithms in software engineering and testing), and EvoSTOC (evolutionary algorithms in stochastic and dynamic environments).

[EN] Diversity of food systems and their interaction with the environment has become a research topic for many years. Scientists use various models to explain environmental issues of food systems. This paper gives an overview of main streams in analyzing this topic. A literature review was performed by analyzing published scientific papers on environmental impacts in the food chain. The selection criteria were focused on different environmental approaches applied in the food chain and on the perspectives of future research. This review shows that on the one side there are generic environmental models developed by environmental scientists and as such applied on food. On the other side, there are models developed by food scientists in order to analyze food-environmental interactions. The environmental research in food industry can be categorized as product, process or system oriented. This study confirmed that the focus of product based approach is mainly performed through life-cycle assessments. The process based approach focuses on food processes such as heat transfer, cleaning and sanitation and various approaches in food waste management. Environmental systems in the food chain were the least investigated stream analyzing levels of environmental practices in place. Future research perspectives are the emerging challenges related to environmental impacts of novel food processing technologies, innovative food packaging and changes in diets and food consumption in connection with climate and environmental changes., The authors would like to acknowledge networking support by the COST Action CA15118 (FoodMC).

Inspiring metaphors play an important role in the beginning of an investigation, but are less important in a mature research field as the real phenomena involved are understood. Nowadays, in evolutionary computation, biological analogies should be taken into consideration only if they deliver significant advantages.

The analysis of worst-case behavior in wireless sensor networks is an extremely difficult task, due to the complex interactions that characterize the dynamics of these systems. In this paper, we present a new methodology for analyzing the performance of routing protocols used in such networks. The approach exploits a stochastic optimization technique, specifically an evolutionary algorithm, to generate a large, yet tractable, set of critical network topologies; such topologies are then used to infer general considerations on the behaviors under analysis. As a case study, we focused on the energy consumption of two well-known ad hoc routing protocols for sensor networks: the multi-hop link quality indicator and the collection tree protocol. The evolutionary algorithm started from a set of randomly generated topologies and iteratively enhanced them, maximizing a measure of “how interesting” such topologies are with respect to the analysis. In the second step, starting from the gathered evidence, we were able to define concrete, protocol-independent topological metrics which correlate well with protocols’ poor performances. Finally, we discovered a causal relation between the presence of cycles in a disconnected network, and abnormal network traffic. Such creative processes were made possible by the availability of a set of meaningful topology examples. Both the proposed methodology and the specific results presented here – that is, the new topological metrics and the causal explanation – can be fruitfully reused in different contexts, even beyond wireless sensor networks.

One of the most notable features of collectible card games is deckbuilding, that is, defining a personalized deck before the real game. Deckbuilding is a challenge that involves a big and rugged search space, with different and unpredictable behaviour after simple card changes and even hidden information. In this paper, we explore the possibility of automated deckbuilding: a genetic algorithm is applied to the task, with the evaluation delegated to a game simulator that tests every potential deck against a varied and representative range of human-made decks. In these preliminary experiments, the approach has proven able to create quite effective decks, a promising result that proves that, even in this challenging environment, evolutionary algorithms can find good solutions.

Genetic Improvement is an evolutionary-based technique. Despite its relatively recent introduction, several successful applications have been already reported in the scientific literature: it has been demonstrated able to modify the code complex programs without modifying their intended behavior; to increase performance with regards to speed, energy consumption or memory use. Some results suggest that it could be also used to correct bugs, restoring the software’s intended functionalities. Given the novelty of the technique, however, instances of Genetic Improvement so far rely upon ad-hoc, language-specific implementations. In this paper, we propose a general framework based on the software engineering’s idea of mutation testing coupled with Genetic Programming, that can be easily adapted to different programming languages and objective. In a preliminary evaluation, the framework efficiently optimizes the code of the md5 hash function in C, Java, and Python.

International audience; PPSN 2016 hosts a total number of 16 tutorials covering a broad range of current research in evolutionary computation. The tutorials range from introductory to advanced and specialized but can all be attended without prior requirements. All PPSN attendees are cordially invited to take this opportunity to learn about ongoing research activities in our field!

International audience; The iterated prisoner’s dilemma is a famous model of cooperation and conflict in game theory. Its origin can be traced back to the Cold War, and countless strategies for playing it have been proposed so far, either designed by hand or automatically generated by computers. In the 2000s, scholars started focusing on adaptive players, that is, able to classify their opponent’s behavior and adopt an effective counter-strategy. The player presented in this paper, pushes such idea even further: it builds a model of the current adversary from scratch, without relying on any pre-defined archetypes, and tweaks it as the game develops using an evolutionary algorithm; at the same time, it exploits the model to lead the game into the most favorable continuation. Models are compact non-deterministic finite state machines; they are extremely efficient in predicting opponents’ replies, without being completely correct by necessity. Experimental results show that such player is able to win several one-toone games against strong opponents taken from the literature, and that it consistently prevails in round-robin tournaments of different sizes.

This short paper contains an extended list of references to diversity preservation methodologies, classified following the taxonomy presented in a previous publication. The list has been updated according to the contributions sent to the workshop "Measuring and Promoting Diversity in Evolutionary Computation", held during the conference GECCO 2016.

Chapitre 11; The use of anti-virus software has become something of an act of faith. A recent study showed that more than 80 % of all personal computers have anti-virus software installed. However, the protection mechanisms in place are far less effective than users would expect. Malware analysis is a classical example of cat-and-mouse game: as new anti-virus techniques are developed, malware authors respond with new ones to thwart analysis. Every day, anti-virus companies analyze thousands of malware that has been collected through honeypots, hence they restrict the research to only already existing viruses. This article describes a novel method for malware obfuscation based an evolutionary opcode generator and a special ad-hoc packer. The results can be used by the security industry to test the ability of their system to react to malware mutations.

Chapitre 5; Several machine learning paradigms have been applied to financial forecasting, attempting to predict the market’s behavior, with the final objective of profiting from trading shares. While anticipating the performance of such a complex system is far from trivial, this issue becomes even harder when the investors do not have large amounts of money available. In this paper, we present an evolutionary portfolio optimizer for the management of small budgets. The expected returns are modeled resorting to Multi-layer Perceptrons, trained on past market data, and the portfolio composition is chosen by approximating the solution to a multi-objective constrained problem. An investment simulator is then used to measure the portfolio performance. The proposed approach is tested on real-world data from Milan stock exchange, exploiting information from January 2000 to June 2010 to train the framework, and data from July 2010 to August 2011 to validate it. The presented tool is finally proven able to obtain a more than satisfying profit for the considered time frame.

In the past decades, different evolutionary optimization methodologies have been proposed by scholars and exploited by practitioners, in a wide range of applications. Each paradigm shows distinctive features, typical advantages, and characteristic disadvantages; however, one single problem is shared by almost all of them: the "lack of speciation". While natural selection favors variations toward greater divergence, in artificial evolution candidate solutions do homologize. Many authors argued that promoting diversity would be beneficial in evolutionary optimization processes, and that it could help avoiding premature convergence on suboptimal solutions. The paper surveys the research in this area up to mid 2010s, it re-orders and re-interprets different methodologies into a single framework, and proposes a novel three-axis taxonomy. Its goal is to provide the reader with a unifying view of the many contributions in this important corpus, allowing comparisons and informed choices. Characteristics of the different techniques are discussed, and similarities are highlighted; practical ways to measure and promote diversity are also suggested. (C) 2015 Elsevier Inc. All rights reserved.

In many optimization processes, the fitness or the considered measure of goodness for the candidate solutions presents uncertainty, that is, it yields different values when repeatedly measured, due to the nature of the evaluation process or the solution itself. This happens quite often in the context of computational intelligence in games, when either bots behave stochastically, or the target game possesses intrinsic random elements, but it shows up also in other problems as long as there is some random component. Thus, it is important to examine the statistical behavior of repeated measurements of performance and, more specifically, the statistical distribution that better fits them. This work analyzes four different problems related to computational intelligence in videogames, where Evolutionary Computation methods have been applied, and the evaluation of each individual is performed by playing the game, and compare them to other problem, neural network optimization, where performance is also a statistical variable. In order to find possible patterns in the statistical behavior of the variables, we track the main features of its distributions, skewness and kurtosis. Contrary to the usual assumption in this kind of problems, we prove that, in general, the values of two features imply that fitness values do not follow a normal distribution; they do present a certain common behavior that changes as evolution proceeds, getting in some cases closer to the standard distribution and in others drifting apart from it. A clear behavior in this case cannot be concluded, other than the fact that the statistical distribution that fitness variables follow is affected by selection in different directions, that parameters vary in a single generation across them, and that, in general, this kind of behavior will have to be taken into account to adequately address uncertainty in fitness in evolutionary algorithms.

Predicting the market’s behavior to profit from trading stocks is far from trivial. Such a task becomes even harder when investors do not have large amounts of money available, and thus cannot influence this complex system in any way. Machine learning paradigms have been already applied to financial forecasting, but usually with no restrictions on the size of the investor’s budget. In this paper, we analyze an evolutionary portfolio optimizer for the management of limited budgets, dissecting each part of the framework, discussing in detail the issues and the motivations that led to the final choices. Expected returns are modeled resorting to artificial neural networks trained on past market data, and the portfolio composition is chosen by approximating the solution to a multi-objective constrained problem. An investment simulator is eventually used to measure the portfolio performance. The proposed approach is tested on real-world data from New York’s, Milan’s and Paris’ stock exchanges, exploiting data from June 2011 to May 2014 to train the framework, and data from June 2014 to July 2015 to validate it. Experimental results demonstrate that the presented tool is able to obtain a more than satisfying profit for the considered time frame.

Graphical abstractDisplay Omitted HighlightsWe used cooperative co-evolutionary to analyse routing in Delay-Tolerant Networks.We applied the method to two urban-scale network scenarios: Venice and San Francisco.We ran extensive experiments on large networks running First Contact protocol.We found groups of strong colluding attackers able to reduce the data delivery rate. In novel forms of the Social Internet of Things, any mobile user within communication range may help routing messages for another user in the network. The resulting message delivery rate depends both on the users' mobility patterns and the message load in the network. This new type of configuration, however, poses new challenges to security, amongst them, assessing the effect that a group of colluding malicious participants can have on the global message delivery rate in such a network is far from trivial. In this work, after modeling such a question as an optimization problem, we are able to find quite interesting results by coupling a network simulator with an evolutionary algorithm. The chosen algorithm is specifically designed to solve problems whose solutions can be decomposed into parts sharing the same structure. We demonstrate the effectiveness of the proposed approach on two medium-sized Delay-Tolerant Networks, realistically simulated in the urban contexts of two cities with very different route topology: Venice and San Francisco. In all experiments, our methodology produces attack patterns that greatly lower network performance with respect to previous studies on the subject, as the evolutionary core is able to exploit the specific weaknesses of each target configuration.