Automatic data classification is commonly performed using supervised learning algorithms that build models to classify new instances. However, labelled data are often scarce and expensive to obtain. Semi-supervised learning (SSL) offers a viable alternative by exploiting a few labelled and many unlabelled instances. Among various SSL approaches, graph-based methods are particularly appealing as they rely on neighbourhood structure rather than explicit parametric distributional assumptions. Nevertheless, their performance heavily depends on graph construction, which requires defining a suitable similarity (or distance) function. Since no single similarity function performs optimally across all datasets, this paper introduces two evolutionary approaches – GEAS and ST-GPAS – to automatically design similarity functions for graph-based SSL. Extensive experiments on 50 public datasets show that the proposed methods yield similarity functions leading to superior transductive learning performance compared with traditional distance-based measures and representative metric-learning algorithms. The improvements, measured in Micro-F1 and Macro-F1, are statistically significant under nonparametric tests. GEAS and ST-GPAS achieve closely matched performance, with small but consistent advantages depending on the supervision regime.

This article investigates the use of machine learning (ML) for classifying legal texts, focusing on the challenges posed by imbalanced class distributions in datasets. In Brazil, where legal processes are numerous and complex, ML could improve judicial efficiency and decision-making speed. However, skewed data distributions make it difficult for standard algorithms to perform well across all classes, often neglecting minority categories. To address this, imbalance learning strategies such as oversampling, undersampling, and hybrid methods are applied to balance data sets and improve classification accuracy. Using real legal data from S & atilde;o Paulo, this study evaluates these techniques' effectiveness in binary classification, providing valuable insights into their applicability in legal settings.

The detection of drone aerial images has become a research hotspot due to the extensive application of drones in various fields. However, the presence of a large number of small objects and complex scenes in the images poses severe challenges to the detection task. As an emerging technology, dynamic neural networks, with their input adaptive adjustment mechanism, provide important theoretical and technical support for solving the problem of detecting small objects in complex scenes of drone aerial images. In this work, we propose Dynamic You Only Look Once Object Detector (Dynamic-YOLO). Firstly, we improve You only look once (YOLO) v9. By leveraging the Convolutional Block Attention Module (CBAM) attention mechanism and Space-to-Depth Convolution (SPD-Conv), we optimise its backbone network to enhance the model’s feature extraction ability. Then, we adjust the structure of its neck network, delete the detection head P5 for large objects, and add the detection head P2 for tiny objects. After that, we combine a dynamic router with the improved YOLOv9 to form a dynamic detector, achieving adaptive processing of images. The experimental results on the Vision Meets Drone (VisDrone) benchmark dataset have demonstrated the excellent performance and flexibility of Dynamic-YOLO. When the Floating Point Operations (FLOPs) increase to 101.5 G, the highest mean Average Precision at Intersection over Union 0.5:0.95 (mAP@0.5:0.95) reaches 32.3% and mAP@0.5 hits 51%, representing a 4.0% and 10% improvement over YOLOv9-M. In addition, by dynamically adjusting the parameters of the router, the computational resource consumption and detection accuracy of the model can be flexibly adjusted, effectively achieving an optimised balance between detection performance and resource costs.

Designing effective architectures for Generative Adversarial Networks (GANs) remains a challenging task due to training instability and the complexity of balancing generator and discriminator performance. While Neural Architecture Search (NAS) has shown promise in automating architecture design, existing NAS-GAN approaches often suffer from limited design flexibility and high computational demands. This paper introduces a gradient-based NAS framework, termed Differentiable Architecture Search for GANs with Rollback Mechanism (RASGAN), aimed at addressing these limitations. RASGAN incorporates a hyperparameter rollback to indirectly optimise evaluation metrics such as the Inception Score (IS) and Fr & eacute;chet Inception Distance (FID), leading to higher-quality generative models. Moreover, the search space integrates lightweight convolutional operations to reduce computational and storage overhead without compromising performance. On unconditional image generation tasks, the proposed method achieves competitive results: on CIFAR-10, RASGAN attains IS = 8.98 and FID = 10.31; on STL-10, IS = 10.55 and FID = 22.37. Compared to existing NAS-GAN methods, the architectures discovered by RASGAN are not only more effective but also significantly more efficient, exhibiting reduced parameter size while maintaining strong generative performance.

Software effort estimation predicts resources needed for a project, including person-hours and costs, and is vital for effective planning and budgeting. This paper compares two grammar-based evolutionary algorithms: grammar-based genetic programming (GGP) and grammatical evolution (GE). Both algorithms are tested on public project datasets and compared with machine learning models such as support vector machines, artificial neural networks, and least-squares linear regression. Results demonstrate that GGP and GE outperform alternative methods across two evaluation metrics, highlighting their effectiveness in estimating software effort.

Phishing is a social engineering attack that captures information by impersonating trusted entities. To detect it, researchers use Machine Learning as a classification task. However, phishing datasets are often imbalanced due to Concept Drift and the semantic nature of attacks. Oversampling, undersampling, and hybrid techniques address this, with hybrids combining both strategies for better results. This study examines the impact of optimizationbased sequencing of sampling algorithms on phishing data and compares it to traditional methods. Results show that optimized sequences improve classifier performance and reduce the effects of imbalance.

Instance hardness measures allow one to characterize and understand why some instances are harder to classify than others in a classification dataset. An instance can be hard to classify for different reasons, such as being in an overlapping region of the classes or a region of poor data representativeness. While there are many instance hardness measures in the related literature, they are mainly concerned with measuring class overlap. This paper also addresses measuring sparsity in a dataset by building a proximity graph from data and extracting some network-based measures from the nodes. Experimentally, we show that some of these measures are effective in characterizing instance hardness and complement the ones from the literature by measuring the density of the regions where the instances are located.

We increasingly integrate technology into our daily activities, and using Machine Learning (ML) algorithms in various domains has become a common practice. However, in crucial sectors where algorithmic decisions significantly impact people's lives, there is a need to scrutinize these decisions more carefully. Using these algorithms in critical areas, such as courtrooms, raises concerns about potential bias and prejudice, directly affecting the justice and partiality of these tools. There is an urge to create algorithms supporting ethical decisions. This paper proposes using data balancing techniques to mitigate the sample bias present in datasets, aiming to make subsequent ML algorithm training more impartial. A version of the ADASYN algorithm is developed, which performs data balancing at both the class level and at the level of protected attributes, enhancing the diversity and representativeness of the protected groups in the datasets. Experimental results show the technique can promote greater fairness in the predictions of different ML models while keeping a good trade-off with overall accuracy.

This systematic review of the literature addresses techniques employed to address the problem of data imbalance. A variety of articles were analyzed, exploring strategies such as under-sampling, oversampling, and their combinations to address asymmetry in class distributions. Sensitive metrics, including recall, precision, and F1 score, emerge as crucial in imbalanced contexts. The studies reveal the challenges in selecting appropriate strategies and underscore the importance of adaptive approaches. Innovative solutions, such as adaptive combinations of techniques and integration with specific algorithms, are discussed. The ongoing need for research to address the specific challenges of data imbalance is highlighted.

The rapid evolution and widespread accessibility of non-invasive medical imaging technologies, exemplified by Magnetic Resonance Imaging (MRI) and Computerized Tomography (CT), are fundamentally reshaping medical decisionmaking paradigms. These sophisticated imaging modalities, capable of extracting high-definition medical images, have emerged as integral components of modern healthcare, facilitating precise diagnostics and treatment planning. The escalating adoption of such technologies, however, has accentuated the need for a nuanced understanding and optimization of the performance and productivity of both medical equipment and the teams operating them, mainly due to the high costs and risks caused by their misuse. This work proposes using univariate analytical models to estimate the number of exams performed per day with machine learning algorithms. For such, different energy-related sensors monitoring 25 magnetic resonance equipment from three different brands were considered. The results of the research reveal a compelling validation of the proposed approach. A notably high Pearson correlation coefficient is observed between the predictions generated by the evaluated models and the real measurements obtained through the Radiology Information System (RIS). This robust correlation emphasizes the accuracy and reliability of the estimation models, validating their potential applicability in real-world healthcare scenarios. Furthermore, the study unveils an intriguing trend that distinguishes the performance of electric current sensors. Thirteen out of the 25 evaluated MRI machines demonstrate superior results when equipped with electric current sensors compared to other sensor types. This nuanced insight not only substantiates the critical role of energy-related sensors in predicting equipment performance but also underscores the importance of tailoring monitoring strategies to the unique characteristics of each machine.

Selecting and optimizing Convolutional Neural Networks (CNNs) has become a very complex task given the number of associated optimizable parameters, as well as the fact that the arrangement of the layers present in a CNN directly influences its performance. Several research areas used automation techniques to construct and optimize these architectures, with Grammatical Evolution (GE) being one of the most promising techniques. Although several works proposed solutions to the problem in question, each adopts its own evaluation strategy (e.g., different datasets, evaluation metrics, hardware infrastructure). This divergence makes it difficult to compare the proposed approaches, and consequently, it is not possible to reach safe conclusions about the performance of the solutions. This work proposes an experimental evaluation of several context-free grammars listed in the literature for constructing and optimizing CNNs architectures. In addition, we included four well-known CNNs as baselines: DenseNet169, EfficientNetB1, InceptionV3 and ResNet50V2. We aim to identify the best practices for elaborating grammars and compare their results with consolidated CNNs for image classification problems in the literature. Besides, we assessed all approaches on the same controlled environment (e.g., datasets, evaluation metrics, software and hardware setup) to guarantee fairness in the evaluation process. The experiments were carried out by investigating the performance of the models generated by different grammars in solving image classification problems in three datasets of variable dimensions: CIFAR-10, EuroSAT, and MNIST. The experiments have validated several key findings: (i) the significance of optimizing Convolutional Neural Networks (CNNs); (ii) the potential of grammar-based methods as a promising alternative for this task, yielding CNN models that outperform state-of-the-art CNN architectures while possessing fewer trainable parameters, resulting in reduced computational complexity; (iii) grammars incorporating regularization layers (such as dropout and batch normalization) and those that confine the search space (via parameter constraints on CNNs) consistently produce high-performing models with lower complexity, even after a few generations of the evolutionary process; and (iv) the selection of the grammar for optimization can positively or negatively impact the model generation, depending on the specific task requirements.

The widespread use of machine learning (ML) in software engineering (SE) encounters a notable challenge: the need for various domain-specific parameters in algorithms. The issue arises when attempting to reuse these parameters across different applications, resulting in sub-optimal outcomes. This hindrance significantly contributes to the limited migration of ML solutions from research labs to industrial settings. This paper underscores the pressing need for novel research to tackle the overarching problem of generic algorithm customisation. To address this, we propose leveraging Automated Machine Learning (AutoML) approaches. These techniques automatically select intelligible models and their corresponding hyper-parameters for forecasting software defect-proneness. More specifically, this paper adapts an AutoML approach to the field of software defect prediction, namely a hyper-heuristic evolutionary algorithm for automatically designing decision tree algorithms (HEAD-DT), originally proposed to address a generic optimisation problem. We benchmark against the popular general software defect-proneness prediction framework (GSDP) and some standard classifiers. Experimental results reveal that the proposed HEAD-DT implementation surpasses other algorithms across three distinct evaluation measures.

The approaches to tune Artificial neural networks (ANN) for running on edge devices, such as weight quantization, knowledge distillation, weight low-rank approximation, and network pruning, usually reduce their accuracy (gap 1). Moreover, they usually require at least 32-bit microcontrollers, leaving out of the equation widely used and much cheaper platforms mostly based on 8-bit microcontrollers (e.g., ATMega328p and ATMega2560), such as Arduino (gap 2). Those microcontrollers can cost between $0.01 to $0.10 on a large scale and can make viable extending IoT applications to a wider range of cheaper personal objects, such as bottles, cans, and cups. In this context, the present study addresses those two identified gaps by proposing and evaluating a technique for tuning ANN to run on 8-bit microcontrollers. 16,000 ANN with distinct configurations were trained and tuned with four widely used datasets and evaluated on two 8-bit microcontrollers. Using less than 3.5Kbytes, the embedded ANN average accuracies outperformed their benchmarks on a 64-bit computer.

The design of complex and deep neural networks is often performed by identifying and combining building blocks and progressively selecting the most promising combination. Neuroevolution automates this process by employing evolutionary algorithms to guide the search. Within this field, grammar-based evolutionary algorithms have been demonstrated to be powerful tools to describe and thus encode complex neural architectures effectively. Following this trend, the present work proposes a novel grammar-based multi-objective neuroevolutionary for generating Fully Convolutional Networks. The proposed method, named Multi-Objective gRammatical Evolution for FUlly convolutional Networks (MOREFUN), includes a new efficient way to encode skip connections, facilitating the description of complex search spaces and the injection of domain knowledge in the search procedure, generation of fully convolutional networks, upsampling of lower-resolution inputs in multi-input layers, usage of multi-objective fitness, and inclusion of data augmentation and optimiser settings to the grammar. Our best networks outperformed previous grammar evolution algorithms, achieving 90.5% accuracy on CIFAR-10 without using transfer learning, ensembles, or test-time data augmentation. Our best models had 13.39±5.25 trainable parameters and the evolutionary process required 90 min per generation.

Most strategies to port ANN on edge devices train large ANN with large datasets on resourceful processors and shrink them (by pruning or using quantization) to fit them on less powerful processors. Those shrunk ANN still usually demand a considerable processor and has lower accuracy. That restricts applications on widely adopted and cheaper 8-bit/16-bit microcontrollers. Also, large datasets make ANN retraining on edge devices unfeasible, shifting it to a cloud environment that makes user experience susceptible to connectivity latencies, restricting some real-time applications. Here, strategies are proposed to train and improve small ANN accuracy by swapping training optimizers and tiny datasets. Around 330,000 ANNs were trained with tiny fractions (<1%) of the MNIST dataset and validated with its complete testing set. Genetic algorithm and tournament heuristics were used to search for the best optimizers combination. Results demonstrated that optimizers combination could improve ANN average accuracy.Moreover, it can achieve similar accuracy with smaller datasets.Among eight optimizers, the top ANNs used most frequently a combination of Adam, Adadelta, and Adamax. This study has potential implications by finding lighter and better ANNs compatible with less powerful edge devices and indicating a research agenda on mixing optimizers for ANN training.

With society’s digitalization, the ever-growing dependence on software increased the negative impact of poor software quality. That impact was estimated at 2.41 trillion to the US economy in 2022. In searching for better tools for supporting quality assurance efforts, such as software testing, many studies have demonstrated the use of Machine Learning (ML) classifiers to predict defective software modules. They could be used as tools to focus test efforts on the potentially defective modules, enhancing the results achieved with limited resources. However, the practical applicability of many of those studies is arguable because of (1) the misuse of their training datasets; (2) the improper metrics used to measure those classifiers’ performance; (3) the use of data from only a system or project; and (4) the use of data from only a computer programing language. When those factors are not considered, the experiments’ results are biased towards a very high accuracy, leading to improper conclusions related to the generalizability of classifiers to practical uses. This study sheds light on those issues and points out promising results by proposing and testing the cross-project and cross-language generalizability of a novel 2-step approach for artificial neural networks (ANN) using a large dataset of 17,147 software modules from 12 projects with distinct programming languages (C, C++, and Java). The results demonstrated that the proposed approach could deal with an imbalanced dataset and outperform a similar ANN trained with the conventional approach. Moreover, the proposed approach was able to improve by 277

Neuroevolution, a sub-field of AutoML, utilizes evolutionary algorithms to automate the process of creating Deep Neural Networks architectures. For problems with complex objects, such as neural networks, Grammar-based Evolutionary Algorithms (GEs) can be used to simplify the implementation and the experimentation by using grammar rules to describe what the components of the complex object are and how they can be connected, that is, they elegantly describe the search space of the problem. In this work, we propose a GE algorithm based on Structured Grammatical Evolution to generate deep convolutional neural networks. Our work has two major contributions: first, the neural networks may contain an arbitrary number and arrangement of skip connections; second, our skip connections may upscale lower-resolution inputs, allowing the generation of architectures such as U-Net. Our best model achieved 0.85 accuracy on CIFAR-10.

Graph representations promise several desirable properties for Genetic Programming (GP); multiple-output programs, natural representations of code reuse and, in many cases, an innate mechanism for neutral drift. Each graph GP technique provides a program representation, genetic operators and overarching evolutionary algorithm. This makes it difficult to identify the individual causes of empirical differences, both between these methods and in comparison to traditional GP. In this work, we empirically study the behavior of Cartesian Genetic Programming (CGP), Linear Genetic Programming (LGP), Evolving Graphs by Graph Programming (EGGP) and traditional GP. By fixing some aspects of the configurations, we study the performance of each graph GP method and GP in combination with three different EAs: generational, steady-state and (1 + λ ). In general, we find that the best choice of representation, genetic operator and evolutionary algorithm depends on the problem domain. Further, we find that graph GP methods, particularly in combination with the (1 + λ ) EA are significantly better on digital circuit synthesis tasks.

Linear Genetic Programming (LGP) represents programs as sequences of instructions and has a Directed Acyclic Graph (DAG) dataflow. The results of instructions are stored in registers that can be used as arguments by other instructions. Instructions that are disconnected from the main part of the program are called noneffective instructions, or structural introns. They also appear in other DAG-based GP approaches like Cartesian Genetic Programming (CGP). This article studies four hypotheses on the role of structural introns: noneffective instructions (1) serve as evolutionary memory, where evolved information is stored and later used in search, (2) preserve population diversity, (3) allow neutral search, where structural introns increase the number of neutral mutations and improve performance, and (4) serve as genetic material to enable program growth. We study different variants of LGP controlling the influence of introns for symbolic regression, classification, and digital circuits problems. We find that there is (1) evolved information in the noneffective instructions that can be reactivated and that (2) structural introns can promote programs with higher effective diversity. However, both effects have no influence on LGP search performance. On the other hand, allowing mutations to not only be applied to effective but also to noneffective instructions (3) increases the rate of neutral mutations and (4) contributes to program growth by making use of the genetic material available as structural introns. This comes along with a significant increase of LGP performance, which makes structural introns important for LGP.

In multi-target prediction, an instance has to be classified along multiple target variables at the same time, where each target represents a category or numerical value. There are several strategies to tackle multi-target prediction problems: the local strategy learns a separate model for each target variable independently, while the global strategy learns a single model for all target variables together. Previous studies suggested that the global strategy should be preferred because (1) learning is more efficient, (2) the learned models are more compact, and (3) it overfits much less than the local strategy, as it is harder to overfit on several targets at the same time than on one target. However, it is not clear whether the global strategy exploits correlations between the targets optimally. In this paper, we investigate whether better results can be obtained by learning multiple multi-target models on several partitions of the targets. To answer this question, we first determined alternative partitions using an exhaustive search strategy and a strategy based on a genetic algorithm, and then compared the results of the global and local strategies against these. We used decision trees and random forests as base models. The results show that it is possible to outperform global and local approaches, but finding a good partition without incurring in overfitting remains a challenging task. Crown Copyright (c) 2021 Published by Elsevier Inc. All rights reserved.