Accurate blood glucose prediction is vital for effective diabetes management, yet the complex, non-linear nature of glycemic dynamics poses a significant challenge for evolutionary regression models. Standard selection mechanisms, which typically rely on aggregated fitness metrics (e.g., global RMSE), often suffer from premature convergence, leading to the stagnation of the search process in local optima. This paper investigates the application of Desire Driven Selection (DDS) within Structured Grammatical Evolution (SGE) for glucose forecasting. Inspired by biological sexual selection, DDS decouples reproduction from strict survival pressure. It allows individuals to co-evolve mating preferences, selecting partners based on "ornaments" that represent performance during specific time-of-day windows rather than a single global score. We evaluated this approach using real-world clinical data from 10 patients with diabetes. Experimental results indicate that SGE+DDS significantly enhances the search capability of the algorithm, outperforming standard Tournament Selection in generalization ability for 9 out of 10 patients. Notably, the proposed method achieved reductions in prediction error of up to 19.5% compared to the baseline on the test data. Analysis reveals that while DDS introduces higher variance between runs compared to the Tournament selection, it increased exploratory capacity, discovering high-quality models that capture difficult glycemic fluctuations.

Image tampering detection faces persistent challenges in localizing manipulation boundaries across diverse forgery techniques, from traditional manipulation to AI-driven diffusion-based synthesis. Both manipulation types exhibit subtle boundary inconsistencies that demand robust feature extraction. However, existing attention mechanisms widely adopted in tampering detection suffer from limitations. Spatial attention produces diluted activations on weak traces, while channel attention amplifies noise from heterogeneous forensic patterns. To refine these widely-used attention mechanisms, we propose BiRNet, a Bilateral-Attentive Refinement Network featuring two key innovations: (1) a Bilateral Cross-Attention Module that enables bidirectional interaction between complementary features with multi-level fusion, and (2) an Attentive Refinement Module that enhances discrimination through parallel local-global pathways. We conduct extensive experiments across seven benchmarks spanning traditional datasets and diffusion-based manipulations. BiRNet achieves an average F1-score of 0.618 across all datasets. These results validate strong cross-domain generalization from conventional to AI-generated forgeries without requiring AIGC-specific training.

• Jointly optimizes region partitioning and region-specific power pool design. • Proposes a dual-agent DRL framework for cooperative region and power decisions. • Reduces action space complexity using modular base-station agents under CTDE. • Demonstrates superior energy efficiency and SIC success compared to fixed or gap-based baselines and outperforms single-agent in convergence. Achieving energy-efficient transmission with high decoding reliability is a fundamental challenge for massive machine-type communication (mMTC) using grant-free Non-Orthogonal Multiple Access (NOMA), due to dense device activity, sporadic traffic, and strong uplink interference. This paper introduces a Centralized Dual-Agent Deep Reinforcement Learning (CDA-DRL) framework that jointly optimizes region partitioning and transmit power pool design in uplink grant-free NOMA. Two cooperative agents at the base station independently learn the number of spatial regions and the number of power levels, respectively, using recurrent Deep Q-Networks under a centralized training and decentralized execution paradigm. This factorized architecture reduces the action-space complexity and enables scalable learning. Simulation results demonstrate that CDA-DRL achieves more stable training, higher Successive Interference Cancellation (SIC) decoding success, and significantly improved energy efficiency, outperforming geometric gap–based baselines by up to 114% and fixed-power schemes by 33%.

The increasing deployment of Artificial Intelligence across various domains has led to a significant rise in power consumption, raising environmental concerns, and highlighting the need for energy-efficient algorithms and hardware. Machine Learning models – particularly Deep Convolutional Neural Networks and Large Language Models – demand substantial computational resources, contributing to higher carbon emissions and reduced sustainability. This work tackles the issue of energy consumption in Machine Learning, with a specific focus on inference. The proposed methodology leverages Neuroevolution to construct effective models while minimizing power usage. This work proposes a novel approach that trains two models simultaneously in a single process, explicitly encouraging one to consume less power without substantially compromising accuracy. It also proposes a mutation strategy that reinserts layer modules with a preference for power-efficient components. This approach is validated in two scenarios using the Fashion-MNIST, CIFAR-10, and CIFAR-100 datasets: (i) evolving models from scratch, and (ii) optimizing pre-trained models for energy efficiency. When evolving from scratch, this method reduces power consumption by up to 49% with only a 2% accuracy drop on Fashion-MNIST, achieves a 20% power reduction on CIFAR-10 while improving accuracy by 0.8%, and enhances accuracy by 12.8% on CIFAR-100 while reducing power usage by 4%. For pre-trained models, this work achieves a 19.8% reduction in power usage on Fashion-MNIST with minimal accuracy loss, a 47% reduction on CIFAR-10 at the cost of a 7.7% drop in accuracy, and a 21.2% power saving on CIFAR-100 despite a 27.2% performance decline.

People with Type 1 diabetes need to predict their blood glucose levels regularly to keep them within a safe range. Accurate predictions help prevent short-term issues like hypoglycemia and reduce the risk of long-term complications. Evolutionary algorithms have shown potential for this task by generating reliable models for glucose prediction.

In nature, survival poses small benefits if one fails to reproduce and spread one's genes. This is particularly relevant in sexually reproductive species, which exerts another pressure dimension on the individual beyond natural selection: Sexual Selection. More often than not, the quality of the chosen mate is a crucial step in reproduction, making all the investment in mate choice worthwhile. This partly explains why partners often prefer certain secondary traits, such as ornaments, particularly if such traits signal good fitness. We hypothesize that the dynamics between mating preferences and fitness-dependent ornaments can act as a filter to find a mate within a population, exploiting good solutions while maintaining high diversity. In this work, we propose a new selection method for Genetic Programming based on these premises, validating our approach on regression problems. Results show that high levels of diversity are maintained when compared against a standard tournament selection with performance gains, reducing the overall error by 16.3% and 13.8% in training and testing respectively, and performing up to par with state-of-the-art Lexicase selection while also providing the best overall solution.

Grammar-Guided Genetic Programming (GGGP) employs a variety of concepts from evolutionary theory to autonomously design solutions for a given task. Recent insights from evolutionary biology can lead to further improvements in GGGP algorithms. In this paper, we propose a new mutation approach called Facilitated Mutation (FM) that is based on the theory of Facilitated Variation. We evaluate the performance of FM on the evolution of neural network optimizers for image classification, a relevant task in Evolutionary Computation, with important implications for the field of Machine Learning. We compare FM and FM combined with crossover (FMX) against a typical mutation approach to assess the benefits of the approach. We find that FMX provides statistical improvements in key metrics, creating a superior optimizer overall (+0.5% average test accuracy), improving the average quality of solutions (+53% average population fitness), and discovering more diverse high-quality behaviors (+523 high-quality solutions discovered on average). Additionally, FM and FMX reduce the number of fitness evaluations in an evolutionary run, reducing computational costs. FM's implementation cost is minimal and the approach is theoretically applicable to any algorithm where genes are associated witha grammar non-terminal, making this approach applicable in many existing GGGP systems.

Decision Trees (DTs) are one of the most widely used supervised Machine Learning algorithms. The algorithm constructs binary tree data structures that partition the data into smaller segments according to different rules. Hence, DTs can be used as a learning process of finding the optimal rules to separate and classify all items of a dataset. Since the algorithm relies on a decision process similar to rule-based decisions, they are easily interpretable. However, DTs can be difficult to analyse when dealing with large datasets and/or with multiple trees, i.e. ensembles. To ease the analysis and validation of these models, we developed a visual tool which includes a set of visualizations that overview and give details of a set of trees. Our tool aims to provide different perspectives over the same data and provide further insights on how decisions are being made. In this article, we overview our design process, present the different visualization models and their iterative validation. We present a use case in the telecommunications domain. In concrete, we use the visual tool to help understand how a model based on DTs decides which is the best channel (i.e., phonecall, e-mail, SMS) to contact a client.

Data is everywhere, our society shapes itself through it and, with the years passing by, it is becoming greater in dimension. A lot of this data now available is multivariate in nature. The bigger the volume and complexity are, the harder tasks to detect, classify, and measure characteristics and relations within data. Glyph-based visualization is one of the possible techniques commonly adopted in data science to address the representation of multivariate data. Multiple varieties of glyph design have been developed and studied over the past decades. However, little research was done to compare the effectiveness of glyphs designs developed for a specific context of the application. This paper aims to study data glyphs and their applications. More specifically, three visual explorations are produced as glyph design alternatives to represent a dataset related to audiological tests carried out in the population of Portugal. These glyphs were evaluated through controlled semi-structured experiments with users, and a crowdsourced experiment, and then an analysis of the performance result of each of the glyphs is presented, evaluating them in terms of learning and memorization. The results show how the use of metaphors and semantic relations to represent the attributes helps in understanding the glyph and its memorization. Additionally, we identified that the redundancy in encoding data might be beneficial.

Abstract Machine learning tools for analysing lubricating oil data have enabled better information in the condition based maintenance (CBM) approach to the process of diagnosing and predicting failures in diesel‐powered vehicle fleets. With the increase in the number of sensors inserted in vehicles, it is possible for companies to stockpile large quantities of information in real time. As such, the development of data‐driven tools will enable accurate identification of the level of wear of a system, evolving CBM into a more reliable and dynamic approach. Following this type of data‐based analysis focusing on determining the wear of systems and equipment, this paper presents an intelligent system for assessing the condition of lubricating oil in automotive diesel engines. To this end, we analyse the use of raw data obtained from the sensors installed in the car and evaluate in conjunction with the insertion of engineered features designed the best way to determine the operating state of the oils. The results presented in this analysis show that to explain 90% of the variation in the original data only the variables kinematic viscosity, dynamic viscosity, engine oil temperature and OSF_v3 are needed. After evaluating the quality of the variables, we conducted an experimental study to analyse the performance of various machine learning algorithms, taking into account the number of features as input data. The results show that the proposed system has the ability to identify the operating conditions of lubricating oil using seven variables as input to a model based on gradient boosting, obtaining a recall result of 93%, precision of 96% and F1‐score of 94%. We conducted a set of additional studies to understand how different subsets of variables affected the performance of the models, and the results show that the best combination includes information regarding the engine speed, coolant and oil temperature, oil pressure, the oil stress factor (OSF_v3), kinematic viscosity, and the dynamic viscosity.

The grammars used in grammar-based Genetic Programming (GP) methods have a significant impact on the quality of the solutions generated since they define the search space by restricting the solutions to its syntax. In this work, we propose Probabilistic Structured Grammatical Evolution (PSGE), a new approach that combines the Structured Grammatical Evolution (SGE) and Probabilistic Grammatical Evolution (PGE) representation variants and mapping mechanisms. The genotype is a set of dynamic lists, one for each non-terminal in the grammar, with each element of the list representing a probability used to select the next Probabilistic Context-Free Grammar (PCFG) derivation rule. PSGE statistically outperformed Grammatical Evolution (GE) on all six benchmark problems studied. In comparison to PGE, PSGE outperformed 4 of the 6 problems analyzed.

NeuroEvolution automates the generation of Artificial Neural Networks through the application of techniques from Evolutionary Computation. The main goal of these approaches is to build models that maximize predictive performance, some-times with an additional objective of minimizing computational complexity. Although the evolved models achieve competitive results performance-wise, their robustness to adversarial examples, which becomes a concern in security-critical scenarios, has received limited attention. In this paper, we evaluate the adversarial robustness of models found by two prominent Neu-roEvolution approaches on the CIFAR-10 image classification task: DENSER and NSGA-Net. Since the models are publicly available, we consider white-box untargeted attacks, where the perturbations are bounded by either the$L_{2}$or the$L_{\infty}$-norm. Similarly to manually-designed networks, our results show that when the evolved models are attacked with iterative methods, their accuracy usually drops to, or close to, zero under both distance metrics. The DENSER model is an exception to this trend, showing some resistance under the$L_{2}$threat model, where its accuracy only drops from 93.70% to 18.10% even with iterative attacks. Additionally, we analyzed the impact of pre-processing applied to the data before the first layer of the network. Our observations suggest that some of these techniques can exacerbate the perturbations added to the original inputs, potentially harming robustness. Thus, this choice should not be neglected when automatically designing networks for applications where adversarial attacks are prone to occur.

This work proposes an extension to Structured Grammatical Evolution (SGE) called Co-evolutionary Probabilistic Structured Grammatical Evolution (Co-PSGE). In Co-PSGE each individual in the population is composed by a grammar and a genotype, which is a list of dynamic lists, each corresponding to a non-terminal of the grammar containing real numbers that correspond to the probability of choosing a derivation rule. Each individual uses its own grammar to map the genotype into a program. During the evolutionary process, both the grammar and the genotype are subject to variation operators.

Building a Machine Learning model requires the use of large amounts of data. Due to privacy and regulatory concerns, these data might be owned by multiple sites and are often not mutually shareable. Our work deals with private learning and inference for the Weighted Random Forest model when data records are vertically distributed among multiple sites. Previous privacy-preserving vertical tree-based frameworks either adapt Secure Multi-party Computation or share intermediate results and are hard to generalize or scale. In contrast, our proposal contains efficient collaborative calculation algorithms of the Gini Index and Entropy for computing the impurity of decision tree nodes while protecting all intermediate values and disclosing minimal information. We offer a learning protocol based on the Paillier Cryptosystem and Digital Envelope. Also, we provide an inference protocol found on the Look-up Table. Our experiments show that the proposed protocols do not cause predictive performance loss while still establishing and utilizing the model within a reasonable time. The results imply that practitioners can overcome the barrier of data sharing and produce random forest models for data-heavy domains with strict privacy requirements, such as Health Prediction, Fraud Detection, and Risk Evaluation.

Supervised machine learning does not hold without data. However, the needed data can be distributed in different locations and are non-shareable under privacy constraints. Methods to circumvent disclosure restrictions in collaborative machine learning are in strong demand. Thus, we propose HERB+ (Homomorphic Encryption for Random forest and gradient Boosting plus), a confidential learning framework for tree-based models under the scenario of vertically dispersed data. While previous related work focused on a specific algorithm, this work presents a wide variety of privacy-preserved and distributed tree-based algorithms (i.e., Decision Tree, Random Forest, and Gradient Boosting Decision Trees for both classification and regression tasks). HERB+ provides the most detailed and general discussions on using Fully Homomorphic Encryption for computing distributed tree-based algorithms during the training process. Our experiments show that although the learning protocols' efficiencies are not optimal, the predictive performance and privacy are preserved. The results imply that practitioners can overcome the barrier of data sharing and produce tree-based models for data-heavy domains with strict privacy requirements, such as Health Prediction, Fraud Detection, and Risk Evaluation.

The error robustness of Convolutional Neural Networks (CNNs) is an important attribute requiring attention due to their growing application in safety-critical domains such as autonomous driving and medical devices. Hardware errors affecting the execution of such models may lead to system failures and, therefore, fault tolerance techniques are necessary to improve dependability. This paper proposes an approach to improve the robustness of CNNs and experimentally compares it with three other existing techniques. Fault injection is used to emulate hardware faults affecting CNNs targeting four distinct datasets. Results indicate that the ranger technique globally provides the best robustness closely followed by the stimulated training technique, although the former provides much lower temporal overhead than the latter. Architectural redundancy and dropout provide varying results. In all cases, caution through final evaluation of any CNN is required, because there are corner cases in which the robustness decreases, contrary to the intended outcome.

Energy is essential for all countries, since it is in the core of social and economic development. Since the industrial revolution, the demand for energy has increased exponentially. It is expected that the energy consumption in the world increases by 50% by 2030 [17]. As such, managing the demand of energy is of the uttermost importance. The development of tools to model and accurately predict the demand of energy is very important to policy makers. In this paper we propose the use of the Structured Grammatical Evolution (SGE) algorithm to evolve models of energy demand, over macro-economic indicators. The proposed SGE is hybridised with a Differential Evolution approach in order to obtain the parameters of the models evolved which better fit the real energy demand. We have tested the performance of the proposed approach in a problem of total energy demand estimation in Spain, where we show that the SGE is able to generate extremely accurate and robust models for the energy prediction within one year time-horizon.

The choice of a proper learning rate is paramount for good Artificial Neural Network training and performance. In the past, one had to rely on experience and trial-and-error to find an adequate learning rate. Presently, a plethora of state of the art automatic methods exist that make the search for a good learning rate easier. While these techniques are effective and have yielded good results over the years, they are general solutions. This means the optimization of learning rate for specific network topologies remains largely unexplored. This work presents AutoLR, a framework that evolves Learning Rate Schedulers for a specific Neural Network Architecture using Structured Grammatical Evolution. The system was used to evolve learning rate policies that were compared with a commonly used baseline value for learning rate. Results show that training performed using certain evolved policies is more efficient than the established baseline and suggest that this approach is a viable means of improving a neural network's performance.

Generative adversarial networks (GANs) achieved relevant advances in the field of generative algorithms, presenting high-quality results mainly in the context of images. However, GANs are hard to train, and several aspects of the model should be previously designed by hand to ensure training success. In this context, evolutionary algorithms such as COEGAN were proposed to solve the challenges in GAN training. Nevertheless, the lack of diversity and premature optimization can be found in some of these solutions. We propose in this paper the application of a quality-diversity algorithm in the evolution of GANs. The solution is based on the Novelty Search with Local Competition (NSLC) algorithm, adapting the concepts used in COEGAN to this new proposal. We compare our proposal with the original COEGAN model and with an alternative version using a global competition approach. The experimental results evidenced that our proposal increases the diversity of the discovered solutions and leverage the performance of the models found by the algorithm. Furthermore, the global competition approach was able to consistently find better models for GANs.

Road injuries are among the top ten causes of death worldwide. It has been shown that providing feedback to drivers decreases the likeliness of having them engaging into dangerous manoeuvres, such as speeding. It also contributes to reduce the amount of life-threatening incidents related with braking. Due to its ubiquity, smartphones are a great resource for assessing driving behaviour. Several mobile applications have been created with this purpose, but there is no concrete evidence that these approaches offer consistent results over distinct platforms (Operating Systems) and hardware. Providing a model for assessing driver behaviour across distinct devices represents a major challenge, due to the increasing differentiation between platforms and mobile devices' internal sensors (gyroscope, accelerometer, GPS, and magnetometer.) In this study we propose the application of Evolutionary Computation techniques to create models for driving behaviour characterisation over data acquired from mobile devices with distinct sensors. Our experiments show that we are able to evolve models that are robust and can accurately identify the legs of a car journey that have abnormal events. In concrete we are able to evolve predictive models that can successfully create a profile about the driving behaviour of a person.

Generative adversarial networks (GAN) present state-of-the-art results in the generation of samples following the distribution of the input dataset. However, GANs are difficult to train, and several aspects of the model should be previously designed by hand. Neuroevolution is a well-known technique used to provide the automatic design of network architectures which was recently expanded to deep neural networks.

Structured grammatical evolution is a recent grammar-based genetic programming variant that tackles the main drawbacks of Grammatical Evolution, by relying on a one-to-one mapping between each gene and a non-terminal symbol of the grammar. It was applied, with success, in previous works with a set of classical benchmarks problems. However, assessing performance on hard real-world problems is still missing. In this paper, we fill in this gap, by analyzing the performance of SGE when generating predictive models for the glucose levels of diabetic patients. Our algorithm uses features that take into account the past glucose values, insulin injections, and the amount of carbohydrate ingested by a patient. The results show that SGE can evolve models that can predict the glucose more accurately when compared with previous grammar-based approaches used for the same problem. Additionally, we also show that the models tend to be more robust, since the behavior in the training and test data is very similar, with a small variance.

Machine learning models are widely adopted in scenarios that directly affect people. The development of software systems based on these models raises societal and legal concerns, as their decisions may lead to the unfair treatment of individuals based on attributes like race or gender. Data preparation is key in any machine learning pipeline, but its effect on fairness is yet to be studied in detail. In this paper, we evaluate how the fairness and effectiveness of the learned models are affected by the removal of the sensitive attribute, the encoding of the categorical attributes, and instance selection methods (including cross-validators and random undersampling). We used the Adult Income and the German Credit Data datasets, which are widely studied and known to have fairness concerns. We applied each data preparation technique individually to analyse the difference in predictive performance and fairness, using statistical parity difference, disparate impact, and the normalised prejudice index. The results show that fairness is affected by transformations made to the training data, particularly in imbalanced datasets. Removing the sensitive attribute is insufficient to eliminate all the unfairness in the predictions, as expected, but it is key to achieve fairer models. Additionally, the standard random undersampling with respect to the true labels is sometimes more prejudicial than performing no random undersampling.

Developing learning systems is challenging in many ways: often there is the need to optimise the learning algorithm structure and parameters, and it is necessary to decide which is the best data representation to use, i.e., we usually have to design features and select the most representative and useful ones. In this work we focus on the later and investigate whether or not it is possible to obtain good performances with compressed versions of the original data, possibly reducing the learning time. The process of compressing the data, i.e., reducing its dimensionality, is typically conducted by someone who has domain knowledge and expertise, and engineers features in a trial-and-error endless cycle. Our goal is to achieve such compressed versions automatically; for that, we use an Evolutionary Algorithm to generate the structure of AutoEncoders. Instead of targeting the reconstruction of the images, we focus on the reconstruction of the mean signal of each class, and therefore the goal is to acquire the most representative characteristics of each class. Results on the MNIST dataset show that the proposed approach can not only reduce the original dataset dimensionality, but the performance of the classifiers over the compressed representation is superior to the performance on the original uncompressed images.

The effectiveness of Artificial Neural Networks (ANNs) depends on a non-trivial manual crafting of their topology and parameters. Typically, practitioners resort to a time consuming methodology of trial-and-error to find and/or adjust the models to solve specific tasks. To minimise this burden one might resort to algorithms for the automatic selection of the most appropriate properties of a given ANN. A remarkable example of such methodologies is Grammar-based Genetic Programming. This work analyses and compares the use of two grammar-based methods, Grammatical Evolution (GE) and Structured Grammatical Evolution (SGE), to automatically design and configure ANNs. The evolved networks are used to tackle several classification datasets. Experimental results show that SGE is able to automatically build better models than GE, and that are competitive with the state of the art, outperforming hand-designed ANNs in all the used benchmarks.

Evolutionary Algorithms are problem solvers inspired by nature. The effectiveness of these methods on a specific task usually depends on a non trivial manual crafting of their main components and settings. Hyper-Heuristics is a recent area of research that aims to overcome this limitation by advocating the automation of the optimization algorithm design task. In this paper, we describe a Grammatical Evolution framework to automatically design evolutionary algorithms to solve the knapsack problem. We focus our attention on the evaluation of solutions that are iteratively generated by the Hyper-Heuristic. When learning optimization strategies, the hyper-method must evaluate promising candidates by executing them. However, running an evolutionary algorithm is an expensive task and the computational budget assigned to the evaluation of solutions must be limited. We present a detailed study that analyses the effect of the learning conditions on the optimization strategies evolved by the Hyper-Heuristic framework. Results show that the computational budget allocation impacts the structure and quality of the learned architectures. We also present experimental results showing that the best learned strategies are competitive with state-of-the-art hand designed algorithms in unseen instances of the knapsack problem.

We present a discrete ant colony algorithm to cluster geometry optimization. To deal with this continuous problem, the optimization framework includes functions to map solutions across the discrete and continuous spaces. Results obtained with short-ranged Morse clusters show that the proposed approach is effective, scalable and is competitive with state-of the-art optimization methods specifically designed to tackle continuous domains. A detailed analysis is presented to help to gain insight into the role played by several components of the ant colony algorithm.