Efficient management of cloud storage resources requires intelligent tier allocation strategies that balance cost optimization with performance requirements. While previous approaches have focused on binary classification schemes for storage tiering, real-world scenarios demand more granular solutions that can adapt to diverse user preferences and workload characteristics. This paper extends our previous work on access frequency prediction by proposing a comprehensive multiclass machine learning framework for intelligent cloud storage tiering. The proposed framework incorporates a novel three-tier classification system ( Cold / Warm / Hot ) and integrates user-centric preferences through a cost-weight parameter, enabling dynamic adaptation to varying preferences along the cost-latency spectrum. We demonstrate the framework’s effectiveness through extensive experiments on real-world access patterns, where we assess the performance of thirteen machine learning algorithms under various user preference profiles. The results show that our multiclass approach achieves cost reductions of up to 40% compared to a static tiering strategy, while providing Pareto-optimal solutions for different user profiles. Through comprehensive Pareto frontier analysis, we demonstrate the framework’s ability to provide transparent trade-off visualization, enabling informed decision-making for cloud storage administrators. Our main contributions are: a multiclass classification approach for storage tiering, the integration of user preferences via parameterized optimization, a comparative analysis of multiple algorithms across different preference configurations, and a practical validation of the framework’s applicability in production cloud storage environments.

Cloud storage providers typically offer multiple tiers with differing performance and cost. Classifying data into correct tiers is challenging, given evolving access patterns. This paper presents a supervised learning framework to predict object access frequency, thus enabling cost-effective tier allocations. Using real-world Dropbox traces, our experiments show up to 37% cost savings compared to an online tiering baseline. We evaluate multiple machine learning methods and time-window strategies, demonstrating the trade-offs between cost optimization and recall (to avoid misclassifving frequently accessed objects).

One of the main events involving the world economy in 2022 was the beginning of the war between Russia and Ukraine. This event offers an opportunity to analyze how a large-magnitude world event can affect the use of cryptocurrencies. Ethereum is one of the most prominent and widely used cryptocurrency platforms and, as such, provides a valuable case study for this scenario. This work investigates the behavior of accounts and their transactions on the Ethereum network during this event. For this purpose, we collect all Ethereum transactions during two distinct periods: (i) during the month the conflict began, and (ii) during the previous year. We organized a dataset with the accounts involved in these transactions and the subset of these accounts that interacted with a service within Ethereum named Flashbots Auction. Flashbots Auction is crucial as it addresses issues regarding transaction ordering and miners exploiting that ordering to make profit. Then, we model temporal graphs in which each vertex represents an account, and each edge represents a transaction between two accounts. We analyzed the behavior of these accounts via graph metrics for both groups during each observed time window. The results show changes in account behavior and activity, as well as variations in daily transaction volume.

Systems Biology is an interdisciplinary field that aims to understand the interactions among biological components. A central focus of this field is modeling gene regulatory networks (GRN) and understanding how gene expression varies. ScRNA -Seq technology has enabled the ability to explore gene expression at the single-cell level, unlike previous technologies where only an average view of gene expression was possible. As a result, the literature has observed a significant increase in the number of inference methods, taking into account the specificities of the data from scRNA -Seq profiling, such as batch effects, biological variations, and dropouts. However, recent studies have shown that the performance of GRN inference algorithms when considering scRNA -Seq technology is close to random predictors. Furthermore, algorithms that perform well on synthetic and curated data are different from those that perform well on experimental data, indicating a lack of robustness. Considering that experimental data is more interesting for biology, as the modeling of its GRNs enables the understanding of biological phenomena, in this paper we show that the CGPGRN framework can deal with experimental data. Computational experiments are carried out and the results indicate that CGPGRN can outperform state-of-the-art algorithms in several situations and is the only one capable of obtaining correct regulatory relationships in all situations considered.

Brain-Computer Interface (BCI) allows direct communication between the human brain and a computer system. Motor Imagery (MI), a prominent BCI paradigm, decodes a mental activity of motor planning in a message to a device. This process can incorporate signal acquisition, preprocessing, temporal and spatial filtering, feature extraction, and classification. Common Spatial Pattern is a well-known method for spatial filtering in BCI. Combining CSP with temporal filtering methods, such as Bandpass and Empirical Mode Decomposition, improves efficiency. This study proposes using CSP in the time-frequency domain with Short-Time Fourier Transform (STFT) as a temporal filter. STFT detects fluctuations in brain activity across time and frequency. It can recognize MI-related elements at various frequencies. We compared STFT-CSP with CSP, Filter Bank CSP, Median-SEE, and Sigmoid-SEE concerning the task of distinguishing between left and right hand movements. The proposed STFT-CSP achieved the best overall result and was more robust than the other approaches. Moreover, by analysing the performance profiles, we can conclude that in the worst-case scenario, the accuracy was improved in more than 30%. Therefore, the method increased result stability across the subjects. The proposed STFT-CSP proved to be more consistent than other approaches and more suitable for real-world applications.

Geopolitical conflicts significantly impact financial networks and systems, e.g., Russia and Ukraine. Cryptoeconomic blockchains such as Bitcoin and Ethereum were introduced as substitutes for traditional financial systems and might behave differently under significant stress. The Russia–Ukraine conflict allowed us to analyze the impact of such complex geopolitical conflicts on the user behaviors of cryptoeconomic blockchains. This article investigates the early stage of such geopolitical conflict using time-varying graphs. We collected and analyzed all the transactions for Bitcoin and Ethereum that took place 2 weeks before and after the conflict started, i.e., we focused on what can be defined as the acute impact of such an event. Our results suggest that the early stage of such geopolitical conflicts may significantly affect cryptoeconomic blockchains’ user behaviors. For instance, we detected that some users behaved more cautiously during the preconflict phase and resumed normalcy during the postconflict phase but exhibited a shift in their behavior. This article analyzes the relationship between the early stages of geopolitical conflicts and cryptoeconomic systems.

Automated Machine Learn (AutoML) process is target of large studies, both from academia and industry. AutoML reduces the demand for data scientists and makes specialists in specific fields able to use Machine Learn (ML) in their domains. An application of ML algorithms is over time-series forecasting, and about these, few works involve the application of AutoML. In this work, an AutoML approach that aggregates time-series forecasting models is proposed. Furthermore, a special focus is given to the optimization stage, which uses genetic algorithm to boost searching for hyper-parameters. In the end, results are compared with a recent time-series forecasting benchmark and we verify that the AutoML model proposed in this work surpasses the benchmark.

Brain-computer interfaces (BCI) are systems that use brain signals to communicate with and control devices, with applications ranging over multiple domains. In healthcare, one of the major applications of BCIs is neurorehabilitation. For example, BCIs help stroke patients recover motor abilities by providing sensory feedback based on imagined movement. Convolutional neural networks (CNN) can be used to classify such motor imagery electroencephalogram (EEG) signals and provide this kind of feedback. However, since these signals are usually noisy and can differ significantly over time and among people, it is frequently necessary to collect a large amount of data to train these models. This process can be time-consuming and fatiguing for the user, impairing the quality of neurorehabilitation treatments and other applications. This paper investigates how data augmentation can mitigate this problem by reducing the need for data and increasing feedback accuracy. We analyze five data augmentation methods from the literature on two motor imagery datasets. We apply data augmentation to a few-parameter CNN in varying settings of EEG electrodes, motor imagery tasks, and number of training samples. Our results show that data augmentation can reduce the amount of original data needed, leading to superior accuracy with 33.33 % fewer training samples in some instances. They also show that combining different data augmentation methods can further improve accuracy.

The Multi-Depot Electric Vehicle Location Routing Problem with Time Windows and Battery Swapping (MDEVLRPTW-BS) aims to mitigate the environmental impacts caused by the use of vehicles with an internal combustion engine and uses economically viable electric vehicles in transportation services with time window restrictions. This problem is complex as one must define (i) the location of the battery swapping stations, (ii) the depots to be used from a list of candidates, and (iii) the assignment and routing of a fleet of electric vehicles. We propose here two constructive methods to solve MDEVLRPTW-BS, namely, Cheapest Insertion and Disruption of Routes. These constructive approaches are used with the Greedy Randomized Adaptive Search Procedure (GRASP) and the Random Variable Neighborhood Descent (RVND) method with six movement operators. Two of these operators, namely Isolate and ChangeBSS, are also proposed here for solving MDEVLRPTW-BS. The proposals are compared with a technique from the literature using instances with 5, 10, and 15 customers, and the proposed approaches presented good results, mainly regarding the processing time. Also, we propose new sets of instances larger than those from the literature. Thus, the search techniques can be analyzed when subjected to greater and more realistic situations. These new instances are composed of 100, 144, 288, 360, 420, and 600 customers. For these cases, the approaches also obtained good and promising results, mainly when the Disruption of Routes approach (GRASP-DR-RVND) is used.

Human activity recognition (HAR) is applicable to a wide range of real-life situations. While machine learning algorithms can be applied for solving this problem, difficulties remain, such as handling a large amount of data available for training and selecting the most appropriate features. Hence, the advent of methods to reduce these issues and improve the currently available algorithms is relevant. Thus, we propose here the application of Cartesian Genetic Programming of Artificial Neural Networks (CGPANN) for training models for HAR. As the computational cost is a relevant issue in this context, high-performance computing strategies in graphic processing units (GPU) are proposed for CGPANN. Two computational experiments are executed and the results show a decrease in computational time spent with the usage of different data structures for the parallel CGPANN on the GPU. Moreover, the CGPANN models for HAR are promising when compared to results from the literature.

The inference of Gene Regulatory Networks (GRNs) from gene expression data is a hard and widely addressed scientific challenge with potential industrial and health-care use. Discrete and continuous models of GRNs are often used (i) to understand the process, and (ii) to predict the values of the relevant variables. Here, we propose a procedure to infer models of GRNs from data where (i) the data is binarized, (ii) a Boolean model is created using a Cartesian Genetic Programming technique, (iii) the obtained Boolean model is converted to a system of ordinary differential equations, and (iv) an Evolution Strategy defines the parameters of the continuous model. As a result, we expect to reduce the effect of noise and to improve biological interpretability. The proposed method is applied to two ODE systems that describe the circadian rhythm network dynamic, with 5 and 10 state variables. The models created by the proposed procedure are able to reproduce the behavior observed in the original data.

The training of machine learning models for imbalanced data classification is a challenging task. Several metrics have been used to assess the performance of the classifiers. Each metric is appropriate for a class of problems, and some users often do not have a clear notion of which metric to use. In such cases, it is desirable that the chosen objective function provides good overall performance for most of the existing metrics. Here, three neuroevolutionary approaches based on Cartesian Genetic Programming are used in order to investigate the impact of optimizing accuracy, G-mean, Fβ-score, and the area under the Receiver Operating Characteristic (ROC) curve when creating classifiers based on Artificial Neural Networks applied to imbalanced data classification problems. The results suggest that the optimization of G-mean and Fβ-score generate models that present a superior overall performance in all metrics.

Brain-Computer Interface (BCI) is an emergent technology with a wide range of applications. For instance, it can be used for post-stroke motor rehabilitation in order to restore part of the motor control of someone injured in an accident. The BCI process involves the signal acquisition and preprocessing of data, extraction and selection of features, and classification. Thus, in order to have a correct classification of the movements, several filters are commonly used to handle the signal data. Here we propose the use of Differential Evolution (DE) with cross-entropy as the objective function to find an appropriate filter. Computational experiments are performed using 2 datasets from BCI competitions for motor imagery with signals of Bipolar and Monopolar Electroencephalography. Also, these problems involve two-class and multiclass classifications. The results show that the proposed DE obtained mean results 9.85% better than the well-known approach Filter Bank Common Spatial Pattern for BCIs with 2 classes for bipolar signals, and it allows for the reduction of the number of electrodes for BCIs with 4 classes.

In order to solve constrained optimization problems, meta-heuristics have to be equipped with a constraint handling scheme. Despite the generality of the meta-heuristics, a large number of objective function (and constraints) evaluations are required so that good results are found. To improve the performance of meta-heuristics it is useful to combine them with exact methods. In this paper, two differential evolution (DE) techniques are proposed to exactly satisfy the linear equality constraints present in a continuous optimization problem that may also include additional non-linear equality and/or inequality constraints. An adaptive penalty method (APM) and a tournament selection technique (TS) are combined to a previous DE variant, called DELEqC-II, to handle the non-linear equality and inequality constraints. Numerical experiments, which include test-problems from the literature, are performed in order to comparatively evaluate the new approaches. The results indicate that the proposed methods outperform the other techniques used in the comparisons.

Video games mimic real-world situations and they can be used as a benchmark to evaluate computational methods in solving different types of problems. Also, machine learning methods are used nowadays to improve the quality of non-player characters in order (i) to create human like behaviors, and (ii) to increase the hardness of the games. Genetic Programming (GP) has presented good results when evolving programs in general. One of the main advantage of GP is the availability of the source-code of its solutions, helping researchers to understand the decision-making process. Also, a formal grammar can be used in order to facilitate the generation of programs in more complex languages (such as Java, C, and Python). Here, we propose the use of Grammar-based Genetic Programming (GGP) to evolve controllers for Mario AI, a popular platform to test video game controllers which simulates the Nintendo's Super Mario Bros. Also, as GP provides the source-code of the solutions, we present and analyze the best program obtained. Finally, GGP is compared to other techniques from the literature and the results show that GGP find good controllers, specially with respect to the scores obtained on higher difficulty levels.

Genetic Programming (GP) is an important technique in evolutionary computing. There has been extensive research and great achievement in GP and its variants. Grammar-based genetic programming (GGP) is one of the most promising ones. We propose here a hybrid approach of GGP with Evolution Strategies (ES). GGP is used to evolve the structure of the models while ES searches for the numerical coefficients in order to improve the overall performance when solving symbolic regression problems. Computational experiments conducted on a set of test-cases reveal that the proposed hybrid approach achieved a good performance when compared to other methods from the literature.

Despite the significance of Artificial Neural Networks (ANNs) in practical situations and the several works available in the literature, to adjust its parameters remains as a current problem. Hence, the advent of methods to assist users during this modeling is relevant. Three hybrid techniques based on Cartesian Genetic Programming (CGP) and Differential Evolution (DE) are proposed here for the construction of ANNs. The developed methods carry out an uncoupled evolution of the topology (using CGP) and the weights (using DE). The ANNs are evolved for classification problems, and seven benchmark datasets are used in the computational experiments. Results show the superiority of the proposed methods when compared to other techniques from the literature.

BACKGROUND: Asthma and allergies prevalence increased in recent decades, being a serious global health problem. They are complex diseases with strong contextual influence, so that the use of advanced machine learning tools such as genetic programming could be important for the understanding the causal mechanisms explaining those conditions. Here, we applied a multiobjective grammar-based genetic programming (MGGP) to a dataset composed by 1047 subjects. The dataset contains information on the environmental, psychosocial, socioeconomics, nutritional and infectious factors collected from participating children. The objective of this work is to generate models that explain the occurrence of asthma, and two markers of allergy: presence of IgE antibody against common allergens, and skin prick test positivity for common allergens (SPT). RESULTS: The average of the accuracies of the models for asthma higher in MGGP than C4.5. IgE were higher in MGGP than in both, logistic regression and C4.5. MGGP had levels of accuracy similar to RF, but unlike RF, MGGP was able to generate models that were easy to interpret. CONCLUSIONS: MGGP has shown that infections, psychosocial, nutritional, hygiene, and socioeconomic factors may be related in such an intricate way, that could be hardly detected using traditional regression based epidemiological techniques. The algorithm MGGP was implemented in c ++ and is available on repository: http://bitbucket.org/ciml-ufjf/ciml-lib .

The Caloric-Restricted Diet Problem (CRDP) aims at finding diets with a reduced caloric count that also respects the nutritional needs of an individual. Thus, it is possible to achieve weight loss without compromising the individual's health. However, due to the small amount of energy contained in such diets, one may not be fully satisfied after a meal. It is possible to overcome this drawback by inserting a larger amount of protein in the diet, as it was shown to be the most effective macronutrient that provides satiety. Thus, this work presents a multi-objective mathematical formulation for the CRDP that minimizes the calorie count of the diet and maximizes the number of proteins ingested. Besides that, a Generalized Differential Evolution algorithm (GDE3) is proposed to solve the resulting problem. Computational experiments are performed with both mono-objective and multi-objective CRDP and two example diets are presented. It shows that it is possible to achieve a diet with a large amount of proteins, while restricting the caloric number.

A differential evolution technique is proposed in order to tackle continuous bilevel optimization problems subject to linear equality constraints, in addition to general non-linear equality and inequality constraints. The idea is to exactly satisfy the linear equality constraints, while the remaining constraints are dealt with via standard constraint handling techniques for metaheuristics. A procedure is proposed in order to generate a random initial population which is feasible with respect to the linear equality constraints. Then a mutation scheme that maintains such feasibility is adopted. The procedure is applied at the lower level (follower) and tested in problems from the literature in order to assess its performance when compared with the case where the constraints are handled via a well known selection scheme for constraint handling. The procedure is applied to problems found in the literature and its performance is compared with the case where the constraints are handled via a well known selection scheme for constraint handling.

Grammatical Evolution (GE) is a genetic programming technique in which the candidate solutions are represented using a binary genotype and the programs can be generated through production rules of a formal grammar. Similarly to other evolutionary computation methods, the GE's performance can be improved when an adequate initial population seeding is adopted. Decision trees are widely used to model classifiers in machine learning and their symbolic form can be mapped back to the GE's binary representation of the candidate individuals. Thus, the use of machine learning techniques to generate decision trees to compose the initial population of GE is investigated here. Computational experiments with a real world data set are carried out and the results show an increase of performance when compared to the traditional seeding approach.

Grammar-based immune programming is proposed here as a tool to assist the search for a general solution to a functional equation. A external archive is incorporated to the algorithm in order to store good solutions found during the search. By inspecting such particular solutions the user is able to generalize and construct a general solution to the functional equation considered. The main objective here is to provide the user with a large diverse set of particular solutions to the problem at hand. Preliminary computational experiments are performed where some functional equations from the literature are tackled.

In recent decades asthma and allergies had great increase worldwide, being currently a serious global health problem. The causes of these disorders are unknown, but the most accepted hypothesis is that improving hygiene and reducing infections may be the main cause of this increase. Both asthma and allergies are complex diseases with strong environmental influence, so the use of versatile tools such as genetic programming can be important in the understanding of those conditions. We applied genetic programming to data obtained from 1296 children. Data related to chronic viral infections and environmental factors were used to classify in asthmatic and non-asthmatic, IgE and SPT in order to assess allergy. For asthma, viral infections were not relevant while for IgE and SPT they were. The use of genetic programming is shown to be a powerful tool to help understand those conditions.

In this work, a genetic algorithm was used to design combinational logic circuits (CLCs), with the goal of minimizing the number of logic elements in the circuit. A new coding for circuits is proposed using a multiplexer (MUX) at the output of the circuit. This MUX divides the truth table into two distinct parts, with the evolution occurring in three sub-circuits connected to the control input and the two data inputs of the MUX. The methodology presented was tested with some benchmark circuits. The results were compared with those obtained using traditional design methods, as well as the results found in other articles, which used different heuristics to design CLCs.

A differential evolution algorithm is proposed here to solve constrained multiobjective optimization problems (CMOPs). In this paper, an Adaptive Penalty Method (APM), which was successfully applied to solve single objective optimization problems, is used to handle the constraints. That constraint handling technique is incorporated to a multiobjective DE which combines the non-dominated ranking and crowding distance schemes when the candidate solutions are replaced. Previously, several variants of the APM were proposed and, here, those variants are tested in order to asses their performance when solving CMOPs. The results obtained in the computational experiments are used to compare the proposal with another well known constraint handling scheme in the literature.

Real world engineering optimization problems are often subject to constraints which are complex implicit functions of the design variables. Frequently, such constrained problems are replaced by unconstrained ones by means of penalty functions. A family of adaptive penalty schemes for steady-state genetic algorithms is proposed here. For each constraint, a penalty parameter is adaptively computed along the run according to information extracted from the current population, such as the existence of feasible individuals and the level of violation of each constraint. The performance of each variant in the family is examined using test problems from the evolutionary computation as well as mechanical and structural optimization literature.

Performance profiles are an analytical tool for the visualization and interpretation of the results of benchmark experiments. In this paper we discuss their explanatory power, and argue that they should be more widely used by the evolutionary computation community. We also introduce some novel performance measures which can be extracted from the performance profiles. In order to illustrate their potential, we apply the referred profiles to the analysis of the results of the CEC 2006 constrained optimization competition. While some of the results are corroborated, some new facts are pointed out and additional conclusions are drawn.

One of the many difficulties that arise in the empirical evaluation of new computational techniques is the analysis and reporting of experiments involving a large number of test-problems and algorithms. The performance profiles are a methodology specifically developed for this purpose which provides a simple means of visualizing and interpreting the results of large-scale benchmarking experiments. However good, performance profiles do not take into account the uncertainty present in most experimental settings. This paper presents an extension of this analytic tool called probabilistic performance profiles. The basic idea is to endow the original performance profiles with a probabilistic interpretation, which makes it possible to represent the expected performance of a stochastic algorithm in a convenient way. The benefits of the new method are demonstrated with data from a real benchmark experiment involving several problems and algorithms.

A genetic algorithm (GA) is hybridized with an artificial immune system (AIS) as an alternative to tackle constrained optimization problems in engineering. The AIS is inspired in the clonal selection principle and is embedded into a standard GA search engine in order to help move the population into the feasible region. The procedure is applied to mechanical engineering problems available in the literature and compared to other alternative techniques.

A genetic algorithm (GA) is hybridized with an artificial immune system (AIS) as an alternative to tackle constrained optimization problems in engineering. The AIS is inspired in the clonal selection principle and is embedded into a standard GA search engine in order to help move the population into the feasible region. The procedure is applied to mechanical engineering problems available in the literature and compared to other alternative techniques.