Keywords: Fuzzy Systems, Intelligent Control
Abstract: This paper presents a sliding mode observer based fuzzy control. The sliding mode observer is developed for a linear dominant system, but taking into account the model mismatch. After that a fuzzy state feedback controller is designed. To ensure the stability of the closed loop system in the presence of uncertainties, Lyapunov synthesis is used. The results are illustrated on a numerical example. Simulations on the nonlinear system are presented to demonstrate the effectiveness of the observer based control.

Keywords: Automated Algorithm, Advanced Optimization, Intelligent Control
Abstract: Engineering processes often require optimizing model variables for satisfactory solutions. Reliable approaches exist in literature but are application-dependent. In that sense, metaheuristics have been proven to deliver outstanding results while imposing a low computing burden. However, choosing the most suitable one from the many available can overwhelm even experts. This study implements a methodology that automatically tailors a problem-based metaheuristic through a hyper-heuristic approach. We select the tuning problem of a Proportional Integral Derivative controller as a case study for achieving the best stable features in an Automatic Voltage Regulator system. The numerical results demonstrate the reliability and potential of the implemented methodology in solving control system tuning. Plus, we conduct an in-depth quantitative comparison with recent works in the literature that support those conclusions.

Keywords: Reinforcement Learning, Intelligent Control, Robotics
Abstract: N/A

Keywords: Robotics, Decision Making, Advanced Optimization
Abstract: A Robotic Mobile Fulfillment System~(RMFS) is a collaborative environment in which a robot delivers products to human for fulfilling orders. However, it is a computationally complex optimization problem. % that integrates diverse optimization problems. In this work, we analyze the feasibility of using high-level solvers for selecting suitable low-level methods. To this end, we generate 111 instances distributed into two datasets. Moreover, we implement two kinds of high-level solvers. The first one is a set of handcrafted rules. The second approach uses a decision tree. % to choose a suitable low-level method for each instance. Our data reveals that it is possible to construct high-level solvers that benefit from the different strengths of the low-level methods by selecting which one to apply. The rules produced by hand and the decision trees high-level solvers are competitive concerning the best individual performer in terms of two standard metrics for this problem: throughput time and orders completed.

Keywords: Intelligent Control, Robotics, Transportation and Vehicle Systems
Abstract: This work proposes the design of a robust controller for the perturbed kinematic model of the unicycle mobile robot, considering a neuro-identifier that imposes restrictions on the identification error. The controller is based on integral sliding modes (ISMs) and the approximation provided by a differential neural network (DNN) for the tracking error dynamics, represented as an uncertain time-varying linear system. The methodology ensures asymptotic stability for the tracking error despite multiplicative disturbances in the control channel. The ISM compensates for the matched dynamics identified with the DNN. Then, a feedback controller based on a Barrier Lyapunov function minimizes the effect of unmatched dynamics while fulfilling state restrictions by solving a set of Linear Matrix Inequalities. Simulation results show the feasibility of the proposed strategy against classical controllers.

Keywords: Intelligent Control, Signal Processing, Bio-inspired
Abstract: Mathematical models are used to represent a vast array of complex processes in engineering, physics, biology, social science, and economics. Model parameters with significant impacts on identification outcomes are ascertained through parametric sensitivity analysis. Authors previously considered an approximation models for systems with uncertain dynamics using a dynamic neural network. The results obtained from studying the problem of predicting the response variable, indicated that this model has a structural flaw. This flaw manifests as an insensitivity of the weight coefficients to external influences, leading to inaccurate predictions. This insensitivity is marked by the minimal contribution of weight coefficient components in the identification process. This paper discusses modifying learning laws to enhance the sensitivity of the weight coefficients to external signals. Through Lyapunov stability analysis, stable algorithms for weight component evolution that minimize identification error were derived.

Keywords: Data Mining, Deep Learning
Abstract: In many medical applications data is a scarce resource and can often only be obtained with invasive surgery. This is for instance the case for physiological cardiovascular data that is necessary to improve the functionality of assistive heart devices. In this work we explore the viability of a GAN architecture to generate cardiovascular data towards enriching a data set obtained in animal testing on which training of future applications can be improved which potentially reduces the need for further animal testing. We evaluate the usefulness of our synthesized data using a downstream task.

Keywords: Data Mining
Abstract: The self-organizing map (SOM) is an unsupervised neural network that uses neuron weight vectors to model training data. Some neurons, called interpolating units, have weight vectors that do not model training data, and instead represent boundaries between emergent data clusters. Interpolating units are useful for distinguishing such clusters using SOM visualizations. However, automatic (non-visual) detection of interpolating units would be advantageous for SOM analysis. This paper proposes a novel algorithm, based on inter-neuron distances in weight vector space, for identifying and possibly pruning interpolating units. Existing methods for interpolating unit detection are surveyed, highlighting drawbacks not associated with the proposed algorithm. Focusing on classification task performance, SOMs which are pruned using the proposed algorithm are compared to unpruned SOMs. This analysis demonstrates that interpolating unit pruning does not adversely affect SOM model quality, and suggests that the proposed algorithm warrants further investigation for application in data science.

Keywords: Data Mining, Decision Making
Abstract: Data mining techniques, prominently clustering, assume a pivotal role in fortifying precision medicine by facilitating the revelation of patient subgroups that share common attributes. By harnessing clustering for the analysis of data behavior within the realm of precision medicine, distinctive disease patterns, and progression dynamics are unveiled, thereby contributing to the formulation of precisely tailored treatment strategies. This paper aims to present the outcomes derived from a clustering analysis applied to diverse clinical datasets encompassing critical facets such as vital signs, laboratory exams, medications, sepsis, Glasgow Coma Scale, procedures, interventions, diagnostics, and admission/discharge records. This compilation of datasets pertains to a cohort of seventy patients. The resultant analysis uncovers intrinsic patterns and relationships residing within intricate datasets. Executed following the rigorous CRISP-DM methodology, this discovery study identified three distinct clusters that group similar data characteristics, encompassing both categorical and numerical clinical data, and resulted in three major groups: patients with stable health conditions, recovery stage, and at risk. This pivotal outcome catalyzes future endeavors, including classification tasks aimed at identifying new patients within specific classes, thereby advancing the horizons of precision medicine.

Keywords: Data Mining, Human-Like Intelligence, Automated Algorithm
Abstract: The surge of interest in artificial intelligence systems has sparked new directions of research in the realm of music data analysis. At the same time, the exploration and exploitation of intrinsic musical structures and sequences, a timeless endeavor, continue to captivate scholars and practitioners alike.

In this context, the fusion of computational techniques with music analysis emerges as a natural progression. One of the pivotal crossroads in this convergence is the identification of musical phrase boundaries, pivotal demarcations that underpin the organizational fabric of a musical composition.

This article pioneers an inventive approach to address the challenge of detecting these musical phrase boundaries, harnessing the power of artificial neural networks. However, the innovation does not stop there; the pinpointed phrase boundaries undergo a comprehensive dissection utilizing pattern mining techniques. The focus of this analysis is on unveiling recurrent motifs and classifying phrases into coherent clusters, predicated on the repetitions and similarities exposed through these neural network-driven techniques.

This exploration was conducted using an extensive repository of folk songs, a treasure trove of foundational musical expressions that indelibly shape the stylistic contours of musical compositions. We claim that the presented approach not only opens avenues for penetrating and nuanced analysis, but also enriches our comprehension of the intricate interplay of musical components and their manifestations inspired by neural networks.

Keywords: Data Mining, Big Data, Evolvable Systems
Abstract: Concept drift detection in noisy data streams is challenging yet essential. This paper introduces NPRDD, a new concept drift detection algorithm that is robust to noise and accurately identifies Real drifts. NPRDD operates on a moving window of recent data, utilizing predicted class probabilities and cross-entropy-based surprise measures to weigh real drift candidates. In line with the Bayesian definition of Real concept drift, NPRDD considers a sample as a drift candidate when the classifier makes an error but is highly confident in its judgment. We evaluate NPRDD on synthetic datasets by varying the noise levels and comparing its performance with other well-established methods. Our results show that NPRDD outperforms other methods regarding ROC-AUC and Accuracy metrics.

Keywords: Deep Learning, Big Data, Dimension Reduction
Abstract: Conventional scalable clustering-based Deep Neural Network (DNN) algorithms cluster linearly separable data, however, non-linearly separable data in the feature space is harder to cluster. This paper proposes a novel Scalable Deep Neural Network Kernelized Literal Fuzzy C-Means (SDnnKLFCM) and Scalable Deep Neural Network Kernelized Random Sampling Iterative Optimization Fuzzy C-Means for Big Data (SDnnKRSIO-FCM). These kernelized clustering methods solve non-linear separable issues by non-linearly transforming the input data space into a high-dimensional feature space using a Cauchy Kernel Function (CKF). We create kernelized deep neural network fuzzy clustering methods using Apache Spark in-memory cluster computing technique to efficiently cluster Big Data on a High-Performance Computing (HPC) machine. To demonstrate the effectiveness of the proposed (SDnnKLFCM) and (SDnnKRSIO-FCM) in comparison to previous scalable deep neural network clustering methods, extensive tests are carried out on a variety of large datasets. The reported experimental results show that the kernelized non-linear deep clustering algorithms in comparison with linear fuzzy clustering algorithms achieve significant improvement in terms of Normalized Mutual Information (NMI), Adjusted Rand Index (ARI), and F-score, respectively.

Keywords: Pattern Recognition, E-health, Ensemble Learning
Abstract: In this study, we focus on Parkinson's Disease (PD) classification and present a comparative analysis of prominent machine learning models using two distinct and independent modalities: Magnetic Resonance Imaging (MRI) and Acoustic data. Unlike many existing works that typically focus on a single modality, our research study provides performance evaluation on the performance of various algorithms on both MRI and Acoustic data. Through a detailed investigation, we provide an understanding of how different models perform when applied to each modality individually. Furthermore, our study extends beyond this comparative framework by introducing an ensemble approach aimed at enhancing the performance of machine learning models for PD classification using the acoustic data. Notably, our ensemble approach yields around a 12% increase in overall performance.

Keywords: Deep Learning, Explainability
Abstract: Early detection of neurodegenerative diseases can be challenging, where Deep Learning (DL) techniques have shown promise. Most DL techniques provide a robust and accurate classification performance. However, due to the complex architectures of the DL models, the classification results are difficult to interpret, causing challenges for their adoption in the healthcare industry. To facilitate this, the current work proposes an effective and interpretable analysis pipeline that compares the performances of pre-trained models for the early detection of Alzheimer's Disease (AD) and Parkinson's Disease (PD). The proposed pipeline allows tuning of hyperparameters, such as batch size, number of epochs, and learning rates, to achieve more robust and accurate classification. Additionally, validation of predictions using heatmaps drawn from GradCAM are also provided.

Keywords: Biometric Systems, Data Mining, E-health
Abstract: Abstract—Gait analysis is becoming increasingly central in various fields such as biomechanics, medicine, and sports. This complex study assesses how individuals walk, weighing in on aspects like bone alignment, joint range of motion, and neuro- muscular activity. Its profound importance emanates from its capability to identify potential walking anomalies at an early stage, which can offset the need for invasive medical interventions and offer valuable insights for clinicians to make well-informed decisions. Inspiration for this in-depth gait analysis research was sourced from previous attempts that employed tools like step mats and cameras to track and analyze movement. Several design imperatives were consolidated within this framework, including the vital need of user safety, assuring comfort throughout usage, preserving system stability, and the undoubtedly significant feature of keeping the insole lightweight. These priorities were not merely for user comfort but were essential for the fidelity of the data being captured. In this paper, it is described the design and optimization processes of an insole for gait analysis following an agile methodology which involves the following stages: requirements, design, implementation and testing. This study delves deeply into the challenges of designing, implement- ing, and optimizing instrumented insoles for gait analysis. The requirements were successfully achieved by creating a 3mm-thick, flexible insole using TPU 95 material through 3D printing. This insole effectively encapsulates electronic components, ensuring comfort, durability, and safety. Index Terms—Gait analysis, insole, unobtrusive wearable

Keywords: Randomized Algorithms, Bio-inspired, Data Mining
Abstract: The deployment of Machine Learning models on hardware devices has motivated a notable research activity around different strategies to alleviate their complexity and size. This is the case of neural architecture search or pruning in Deep Learning. This work places its focus on simplifying randomization-based neural networks by discovering fixed-point quantization policies that optimally balance the trade-off between performance and complexity reduction featured by these models. Specifically, we propose a combinatorial formulation of this problem, which we show to be efficiently solvable by multi- objective evolutionary algorithms. A benchmark for time series forecasting with Echo State Networks over 400 datasets reveals that high compression ratios can be achieved at practically admissible levels of performance degradation, showcasing the utility of the proposed problem formulation to deploy reservoir computing models on resource-constrained hardware devices.

Keywords: Bio-inspired, Advanced Optimization, Decision Making
Abstract: Hopfield networks are an attractive choice for solving many types of computational problems because they provide a biologically plausible mechanism. The Self-Optimization (SO) model adds to the Hopfield network (HN) by using a biologically founded Hebbian learning rule, in combination with repeated network resets to arbitrary initial states, for optimizing its own behavior towards some desirable goal state encoded in the network. However, the solutions to the abstract problems used in the literature offer little insight into how HN arrive at solutions or partial solutions. In order to better understand that process, we demonstrate first that the SO model can solve concrete combinatorial satisfiability problems: The Liars problem and the map coloring problem. Based on these solutions, we discuss how under certain conditions critical information might get lost forever with the learned network producing seemingly optimal solutions that are in fact inappropriate for the problem it was tasked to solve. What appears to be an undesirable side-effect of the SO model, can provide insight into its process for solving intractable problems.

Keywords: Advanced Optimization, Bio-inspired, Human-Like Intelligence
Abstract: Evolutionary algorithm(EA) with knowledge transfer is an emerging research topic that has attracted a lot of attention in the EA community. The multiform evolutionary algorithm is a promising algorithm framework concept falling in this direction, but little work has been done so far to design and discuss this new EA framework. In this paper, we proposed an adaptive multiform evolutionary framework, which integrates the idea of transfer optimization and population-based evolutionary algorithms. The main idea is to utilize multiple equivalent or similar formulations to solve the given problem cooperatively. By adaptively adjusting computational resources of different formulations and transferring knowledge among formulations, the search efficiency and the population diversity can be improved. Furthermore, a new multiform algorithm is implemented based on the proposed framework, which utilizes two different coordinate systems, namely the Cartesian and polar coordinate systems, to model the given problem. To test the effectiveness of the proposed algorithm, we conducted numerical experiments on CEC2013 benchmark test functions, and the experimental results have verified the efficacy of the proposed algorithm.

Keywords: Deep Learning
Abstract: Binary neural networks have emerged as an efficient solution for resource-constrained devices due to their reduced computational, memory, and storage requirements. However, binary neural networks often suffer from decreased accuracy compared to floating-point models. In this study, we propose a binary residual multilayer perceptron architecture that mitigates the degradation caused by binarization through the incorporation of normalization layers and residual connections. By leveraging design recommendations from state-of-the-art binary architectures, we aim to create a user-friendly model that can be easily implemented without requiring extensive neural network design expertise. This paper presents the empirical results of our proposed architecture, demonstrating its effectiveness in reducing degradation and improving performance for hardware-constrained devices.

Keywords: Pattern Recognition, Fault Detection, Advanced Optimization
Abstract: In this paper, we present an adaptation of Newton's method for the optimization of Subspace Support Vector Data Description (S-SVDD). The objective of S-SVDD is to map the original data to a subspace optimized for one-class classification, and the iterative optimization process of data mapping and description in S-SVDD relies on gradient descent. However, gradient descent only utilizes first-order information, which may lead to suboptimal results. To address this limitation, we leverage Newton's method to enhance data mapping and data description for an improved optimization of subspace learning-based one-class classification. By incorporating this auxiliary information, Newton's method offers a more efficient strategy for subspace learning in one-class classification as compared to gradient-based optimization. The paper discusses the limitations of gradient descent and the advantages of using Newton's method in subspace learning for one-class classification tasks. We provide both linear and nonlinear formulations of Newton's method-based optimization for S-SVDD. In our experiments, we explored both the minimization and maximization strategies of the objective. The results demonstrate that the proposed optimization strategy outperforms the gradient-based S-SVDD in most cases.

Keywords: Randomized Algorithms, Bio-inspired, Advanced Optimization
Abstract: Runtime results of multiobjective evolutionary algorithms in unbounded integer spaces are scarce at present. In order to advance this research field we consider two versions of the (1+1)-EA and analyze their runtime to the Pareto front of a carefully designed biobjective test problem.

Keywords: Explainability, Bio-inspired, Advanced Optimization
Abstract: The quality of individuals in evolutionary algorithms (EAs) is usually measured in terms of their fitness. If an individual has a good fitness, a good genome is assumed. However, a good fitness value does not guarantee that the individual can produce good offspring and guide the algorithm towards the global optimum. Answering the question of what makes a genome good is not trivial, especially when considering different types of crossover operators, copying or combining genome values. This work aims towards answering this question by evaluating the influence of optimal gene values in the initial population of EAs. In computational experiments, a random population is seeded with generated individuals of different fitness qualities and containing different amounts of optimal genetic material. Tests are done for multiple dimensions and with crossover operators copying or combining the parents genes to the offspring. Data is evaluated both in terms of algorithmic performance and population dynamics, clearly showing the influence of optimal gene values.

Keywords: Bio-inspired
Abstract: Self-Driving Automata (SDAs) are variations on finite automata that both read and output symbols. They are versatile and practical when used for the generation of data for a variety of problems. In this study, we examine several questions regarding their operation, using sequence matching as a test problem in the analysis. We present a new mutation operator and four dynamic mutation adjusters. We analyze these, along with crossover, for their ability to solve the problem and their relative ability to improve the population; in all of these, we also examine population diversity over time. We find that using mutation that implements a static quantity of changes outperforms one with dynamic changes. Further, while population diversity does decrease somewhat, evolution is still possible.

Keywords: Evolving Learning, Randomized Algorithms, Advanced Optimization
Abstract: Population initialization is a crucial and essential step in evolutionary multi-objective optimization (EMO) algorithms. The quality of the generated initial population can significantly affect the performance of an EMO algorithm. However, few studies have focused on designing a generalized initialization method to improve the performance of EMO algorithms in solving multi-objective combinatorial optimization (MOCO) problems. Most of the existing advanced initialization methods involve complex techniques tailored to the specific characteristics of the problems to be solved. In this paper, we propose a general and effective framework of population initialization for EMO algorithms, aiming to improve their performances in solving various MOCO problems. Our approach involves the inclusion of a few specific heuristic solutions, including extreme solutions and a center solution, into the initial population. This inclusion serves to guide the evolution of the population throughout the optimization process. Our experimental results show that initial populations with a few heuristic solutions significantly improve the performance of EMO algorithms. Algorithm behavior analysis and further study are also provided, allowing for a comprehensive understanding of the effectiveness and applicability of our proposed method.

Keywords: Reinforcement Learning, Deep Learning, Data Mining
Abstract: The promising results of machine learning in time series classification, along with the rise in sensor data-driven use cases, have led to the increasing deployment of models in IoT environments, on edge devices. Since these devices are typically resource constrained, they cannot always execute large and complex models, so they often offload (part of) their tasks to remotely located models. This synergy however introduces the need to transfer a large amount of sensor data to the cloud, which can be detrimental to bandwidth cost and inference speed of the application, and energy utilization of the device. Although techniques such as early classification can limit the data that has to be transferred, there are still unexplored opportunities when it comes to input filtering. A recent versatile early-exit framework, extending early classification and adapting it to multivariate time series, has investigated this potential. In this work, we propose a variation of this method, creating a more flexible, reinforcement learning-enabled framework that can adapt the input variables (channels) considered for classification across time, aiming for maximizing accuracy while minimizing the input data necessary. Extensive testing on synthetic data and real datasets shows that our method can, in multiple cases, achieve better accuracy for a similar percentage of input filtering, both compared to the baseline framework, as well as to the conventional early classification approach.

Keywords: Reinforcement Learning, Agent-Based Modeling, Robotics
Abstract: Exploration poses a fundamental challenge in Reinforcement Learning (RL) with sparse rewards, limiting an agent's ability to learn optimal decision-making due to a lack of informative feedback signals. Self-Imitation Learning (self-IL) has emerged as a promising approach for exploration, leveraging a replay buffer to store and reproduce successful behaviors. However, traditional self-IL methods, which rely on high-return transitions and assume singleton environments, face challenges in generalization, especially in procedurally-generated (PCG) environments. Therefore, new self-IL methods have been proposed to rank which experiences to persist, but they replay transitions uniformly regardless of their significance, and do not address the diversity of the stored demonstrations. In this work, we propose tailored self-IL sampling strategies by prioritizing transitions in different ways and extending prioritization techniques to PCG environments. We also address diversity loss through modifications to counteract the impact of generalization requirements and bias introduced by prioritization techniques. Our experimental analysis, conducted over three PCG sparse reward environments, including MiniGrid and ProcGen, highlights the benefits of our proposed modifications, achieving a new state-of-the-art performance in the MiniGrid-MultiRoom-N12-S10 environment.

Keywords: Reinforcement Learning, Autonomous Systems, Fuzzy Systems
Abstract: What indications are there when the environment changes and the learned policy is no longer optimal? Is it possible to predict when a non-stationary environment changes in some way? In this paper we propose a method that helps agents know when to retrain their policies via reinforcement learning. The agents detect changes based on the temporal difference. A hierarchical learning model is used to aid in these non- stationary environments. The hierarchical learning model has two levels, the higher-level policy, which we call the learning switch, and the lower-level policy, which tells the agents their suitable action to play the game. The higher-level policy determines when reinforcement learning should be turned on or off based on the temporal differences calculated within a game or episode. Two multi-agent differential games are used as examples. The first two examples tackle the problem in cooperative games, while the last example addresses the competitive game. The results show that the agents can maintain suitable performance by switching on the learning process for a few iterations after environment changes occurs.

Keywords: Reinforcement Learning, Decision Making, Intelligent Control
Abstract: In complex real-world decision problems, ensuring safety and addressing uncertainties are crucial aspects. In this work, we present an uncertainty-aware Reinforcement Learning agent designed for risk-sensitive applications in continuous action spaces. Our method quantifies and leverages both epistemic and aleatoric uncertainties to enhance agent's learning and to incorporate risk assessment into decision-making processes. We conduct numerical experiments to evaluate our work on a modified version of Lunar Lander with variable and risky landing conditions. We show that our method outperforms both Deep Deterministic Policy Gradient (DDPG) and TD3 algorithms by reducing collisions and having significant faster training. In addition, it enables the trained agent to learn a risk-sensitive policy that balances performance and risk based on a specific level of sensitivity to risk required for the task.

Keywords: Reinforcement Learning, Robotics
Abstract: Meta-reinforcement learning aims to overcome important limitations in reinforcement learning, like low sample efficiency and poor generalization, by creating agents that adapt to new tasks. The development of intelligent robots would benefit from such agents. Long-standing issues like data collection and generalization to real-world dynamic environments could be mitigated by sample-efficient adaptable algorithms. However, most such algorithms have only been proven to work in low-complexity environments. These provide no guarantee that a near-optimal global policy does not exist, which makes it difficult to evaluate adaptable policies. This hinders the in-depth analysis of an agent's potential to adapt, while also introducing a gap between controlled experiments and real-world applications. We propose MEWA, a collection of task distributions used as a benchmark for adaptable agents. Our tasks contain a shared structure that an agent can leverage to learn the task-specific structure of new tasks. To ensure our environment is adaptive, we select some of the task parameters using the solution to a constrained optimization problem. Other parameters are randomized, allowing the creation of arbitrary task distributions. We evaluate three state-of-the-art meta-reinforcement learning algorithms on our benchmark, that were previously shown to adapt to new tasks with a simpler structure. Results show that the algorithms can reach meaningful performance on the task, but cannot yet fully adapt to the task-specific structure. We believe this benchmark will help identify some of the issues that hinder adaptability, ultimately aiding in the design of new algorithms, more suitable for real-world human-robot applications.

Keywords: Reinforcement Learning, Particle Swarm Optimization, Agent-Based Modeling
Abstract: In engineering processes that require continuous control, it is common to face significant challenges. Addressing these challenges through explicit modeling can take much work and effort. For this reason, Reinforcement Learning (RL) has gained popularity as a feasible strategy for solving this problem. In this context, various value-based methodologies, policies, or combinations have been employed to obtain an optimal learning policy. However, problems such as convergence to local maxima and high variance in training persist. In addition, computational time and cost increase in complex environments, so more robust RL methodologies are required. This paper proposes a Swarm Intelligence Guided Neural Reinforcement Learning (SIGNRL) algorithm, which uses Particle Swarm Optimization as a multi-agent parameter explorer to find the optimal policy. Numerical results obtained in the OpenAI Gym Cart-Pole environment show that SIGNRL, with its gradient-free learning, exhibits good convergence and lower variance in continuous control tasks.

Keywords: Decision Making, Bio-inspired, Swarm Intelligence
Abstract: This article proposes an Ant Colony Optimization (ACO) algorithm, an optimization method to find paths in graphs, adapted to solve strategic games. The games of study are Tic-Tac-Toe (also known as noughts and crosses, three in a row, or Xs and Os), and Chess. The algorithms' performance is contrasted by contending ACO against the Minimax algorithm, in different setups of Tic-Tac-Toe and Chess. The performance is explained in terms of average time response, correctness of the move choice, and memory used when executing the function. Results reveal a slightly better average performance by the ACO algorithm compared to Minimax. These findings highlight the ability of ACO in decision-making algorithms without requiring knowledge of previous games. Furthermore, the results suggest that the ACO-based path optimization approach can be an effective alternative to improve the efficiency of decisions made by intelligent systems in environments that require rapid response.

Keywords: Explainability, Deep Learning, Bio-inspired
Abstract: Deep Convolutional Neural Networks (ConvNets) have demonstrated successful implementations in various vision tasks, including image classification, segmentation, and image captioning. Despite their achievements, concerns persist regard- ing the explainability of these models, often referred to as black- box classifiers. While some interpretability papers suggest the existence of object detectors in ConvNets, others refute this notion. In this paper, we address the challenge of identifying such neurons by utilizing Hebbian learning to discover the most associated neurons for a given stimulus. Our method focuses on the VGG19 and ResNet50 networks with the Dogs- vs-Cats dataset. During experimentation, we found that the most associated hidden neurons to the labels are not object detectors. Instead, they seem to encode relevant aspects of the category. By shedding light on these findings, we aim to improve the understanding and interpretability of deep ConvNets for future advancements in the field of computer vision.

Keywords: Image Processing, Deep Learning, Pattern Recognition
Abstract: In the era of big data, we now live on, there is an increasing demand to convert large amounts of scanned documents, such as texts, medical records, and images into digital formats. However, often when scanning introduces imperfections such as salt-and-pepper or background noise, blurring caused by camera motion, watermarking, coffee stains, wrinkles, or faded text. These imperfections carry significant challenges to current algorithms of text recognition, leading to a decline in their performance. To date, a wide range of methods are aimed at reducing noise. This work compares the performance of a CycleGAN model concerning median filter, Wiener filter, adaptive threshold, morphological filtering, and a CNN-based autoencoder. While the CNN-based autoencoder technique gave us the best results, the CycleGAN model approach provided us with comparable results with only 50 training epochs in contrast to the 700 epochs of the CNN-based autoencoder and was superior to the rest of the other contrasted methods. Likewise, data preparation for the training is much simpler in the CycleGAN model due to its property of requiring only unpaired data for training.

Keywords: Human-Like Intelligence, Human-Computer Interactions, Robotics
Abstract: A chatbot is a machine with conversational capabilities that tries to resemble a person. In the 90s the A.L.I.C.E. chatbot was created, showing significant advances over its predecessors. Since then, different progress has been made until, thanks to the advancement of technology, the development of current improved models has been achieved. However, now special attention is being paid to chatbots having affective recognition capabilities to enrich the user experience, which is an understudied area. This paper overviews state of the art works on recognition of speaker’s emotions and intentions and proposes to design a speech acts-based model of a chatbot that can interpret human emotions in text and give a coherent response in content and the expressed feelings. A set of techniques will be included in the design to recognize both the user’s emotions and her intentions when expressing herself.

Keywords: Human-Computer Interactions, Deep Learning, Signal Processing
Abstract: Emotion Recognition is a research area that has had a surge in interest, since areas such as mental health, psychological diagnosis, e-learning and assistance for people who are not capable of communicating their feelings, depend on certain level, on the capacities of computer systems to reliably identify emotions. There are several approaches to this task, for instance, analyzing facial expressions, speech, and physiological signals (electrocardiogram, galvanic skin response, electroencephalogram, among others). Electroencephalogram is one of the preferred methods due, in part, to is great temporal resolution. Therefore, in this paper we used the EEG Brainwave Dataset as benchmark to test our model, which is a four layer, one dimensional convolutional neural network. After the preprocessing pipeline, consisting on considering certain features of the dataset as signals and processing them accordingly, by creating several channels by two decomposition methods, our model achieved accuracy values of 98.36% and 95.31%, which is competitive with what is found on the state of the art, while being a significantly less complex model.

Keywords: Data Mining, E-health, Pattern Recognition
Abstract: According to Ministry of Health in Mexico, Hypertension, commonly referred to as High Blood Pressure (HBP), continues to rank among the foremost ten causes of mortality in Mexico. This document describes the methodology for unearthing correlations between non-clinical variables and HBP, utilizing data clustering techniques in a data set derived from diverse Mexican institutions

Keywords: Decision Making, Pattern Recognition, Advanced Optimization
Abstract: Incomplete or sparse non-dominated fronts are unavoidable in multi-objective optimization due to complexity of problems, morphology of Pareto optimal fronts, and stochasticity involved in evolutionary optimization algorithms. It is pragmatic to develop methods that can alleviate some of these issues after the optimization run is complete, without the need for re-optimization or additional solution evaluations. Previously developed methods demonstrated that it is possible to predict Pareto-optimal solutions from pseudo-weight vectors using Gaussian Process Regression (GPR) models. We extend the GPR-based method to predict new Pareto-optimal solutions using reference vectors as unique identifiers and demonstrate that like the pseudo-weight vectors, reference vectors can also used instead in learning the association between identifiers and corresponding variable vectors. Results on many test problems indicate that the choice of a suitable identifier makes a large impact on the decision-making process, particularly for visualizing the newly created non-dominated (ND) solutions. In this study, we discuss the advantages and disadvantages of using pseudo-weights and reference vectors as unique identifiers for ND solutions, paving the way to devise further identifiers for predicting new Pareto-optimal solutions.

Keywords: Explainability, Decision Making
Abstract: This paper is aimed at enhancing the success rate of Genetic Programming (GP) algorithms for symbolic regressions. It is shown that the outcome of GP algorithms over several runs can lead to an optimal solution for such problems, but the success rate, i.e., the number of successful runs, is sometimes small. We address this issue by proposing multi-objective and island model (IM) optimization for GP. We study the influence of various objective functions and IM configurations on the success rates and present 36 algorithm variants, which are tasked with solving two benchmark equations from the fluid mechanics area. This specific benchmark problem has been previously shown to suffer from a low success rate and high variations between the results of multiple runs. Our experiments show a strong influence of the objective functions on the success rate. The additional IM implementation improves the results for some objectives. The algorithm with the highest success rate on the more complex benchmark problem employs both, multiple objectives and IM.

Keywords: Decision Making
Abstract: As objectives increase in many-objective optimization (MaOO), often so do the number of non-dominated solutions, potentially resulting in solution sets with thousands of non-dominated solutions. Such a larger final solution set increases difficulty in visualization and decision-making. This raises the question: how can we reduce this large solution set to a more manageable size? In this paper, we present a new objective archive management (OAM) strategy that performs post-optimization solution set reduction to help the end-user make an informed decision without requiring expert knowledge of the field of MaOO. We create separate archives for each objective, selecting solutions based on their fitness as well as diversity criteria in both the objective and variable space. We can then look for solutions that belong to more than one archive to create a reduced final solution set. We apply OAM to NSGA-II and compare our approach to environmental selection finding that the obtained solution set has better hypervolume and spread. Furthermore, we compare results found by OAM-NSGA-II to NSGA-III and get competitive results. Additionally, we apply OAM to reduce the solutions found by NSGA-III and find that the selected solutions perform well in terms of overall fitness, successfully reducing the number of solutions.

Keywords: Decision Making, Fuzzy Systems
Abstract: A fundamental principle of Granular Computing is that of an information granularity distribution and its optimization process. It has been used in system modelling to elevate a numerical model to its granular counterpart, which is more in rapport with reality. For example, in decision-making with fuzzy preference relations, it has been applied to enhance the existing numerical consistency improvement procedures. However, even though different protocols of information granularity distribution have been proposed, only the one based on a uniform and symmetric distribution and the one based on a symmetric but non-uniform distribution have been considered. Given that there exist others, this study aims to analyze how we can take advantage of all of them to improve the consistency of the fuzzy preference relations. Some numerical experiments are also completed to show the performance of these protocols

Keywords: Randomized Algorithms, Advanced Optimization
Abstract: The hypervolume-based multi-objective evolutionary algorithms (HV-MOEAs) have proven to be highly effective in solving multi-objective optimization problems. However, the computation time of the hypervolume calculation increases significantly as the number of objectives increases. To address this issue, an R2-based hypervolume contribution approximation (R2-HVC) method was proposed. Nevertheless, the original R2-HVC generates a large number of vectors and computes the HVC only once. In this study, we propose an ensemble method based on the R2-HVC method. By using a small number of vectors for repetitive computation and majority voting, the ensemble method can reduce the probability of making incorrect choices. Experimental results show that the proposed method can improve the approximation accuracy while maintaining a similar computation time to the original R2-HVC method.

Keywords: Decision Making, Advanced Optimization
Abstract: In the context of design, multi-concept optimization (MCO) refers to the task of concurrently identifying the best concept and the corresponding variable values to optimize certain objective(s). Despite its relevance in various practical domains such as engineering, transport, and product design, there have been limited studies on developing computationally efficient algorithms specialized for MCO problems. One of the contributing factors towards this gap is the lack of benchmark problems for MCO that offer diverse challenges for the systematic evaluation and development of advanced algorithms. In this paper, we conduct a brief review of some existing multi-objective test problems in the domain and discuss some of their shortcomings. The key aim is to highlight the need for the development of a more extensive set of benchmark problems that are flexible and tunable in terms of the challenges posed to the solution methodologies. In turn, we hope that this will encourage the development of more advanced algorithms to solve practical MCO problems in the future.

Keywords: Evolving Learning, Reinforcement Learning, Robotics
Abstract: The main question this paper addresses is: What combination of a robot controller and a learning method should be used, if the morphology of the learning robot is not known in advance? Our interest is rooted in the context of morphologically evolving modular robots, but the question is also relevant in general, for system designers interested in widely applicable solutions. We perform an experimental comparison of three controller-and-learner combinations: one approach where controllers are based on modelling animal locomotion (Central Pattern Generators, CPG) and the learner is an evolutionary algorithm, a completely different method using Reinforcement Learning (RL) with a neural network controller architecture, and a combination `in-between' where controllers are neural networks and the learner is an evolutionary algorithm. We apply these three combinations to a test suite of modular robots and compare their efficacy, efficiency, and robustness. Surprisingly, the usual CPG-based and RL-based options are outperformed by the in-between combination that is more robust and efficient than the other two setups.

Keywords: Evolvable Systems, Multi-Agent System, Swarm Intelligence
Abstract: Behavioral diversity is known to benefit problem solving in biological social systems such as insect colonies and human societies, as well as in artificial distributed systems including large-scale software and swarm-robotics systems. We investigate methods of evolving robot swarms in which individuals have heterogeneous behaviours. Two approaches are investigated to create swarm of size n. The first encodes a repertoire of n behaviours on a single individual, and hence evolves the swarm directly. The second approach uses two phases. First, a large repertoire of diverse behaviours is evolved and then another evolutionary algorithm is used to search for an optimal allocation of behaviours to the swarm. Results indicate that the two phase approach of generate then allocate produces significantly more effective collective behaviors (in terms of task accomplishment) than the direct evolution of behaviorally heterogeneous swarms.

Keywords: Evolvable Systems, Evolving Learning, Explainability
Abstract: Evolving Fuzzy Neural Networks (EFNNs) are well-regarded for their interpretability and proficiency in pattern classification tasks. However, their accuracy may need to be improved when confronted with limited samples for specific classes or the emergence of new classes in the data stream. To overcome this limitation, we applied the EFNN-Gen, a novel approach that integrates a priori knowledge through generalist rules to solve a beer classification problem. These rules are derived from assessing the specificity of Gaussian functions within the first layer neurons of the EFNN. They represent expert knowledge about the classification problem and are aimed at enhancing the network's performance. Experimental tests conducted on the Beer dataset, a real-world multiclass pattern classification dataset, demonstrate that integrating generalist rules leads to a significant accuracy improvement of 97.14%.

Keywords: Deep Learning, Evolving Learning, Data Mining
Abstract: Training Artificial Neural Networks (ANNs) poses a challenging and critical problem in machine learning. Despite the effectiveness of gradient-based learning methods, such as Stochastic Gradient Descent (SGD), in training neural networks, they do have several limitations. For instance, they require differentiable activation functions, and cannot optimize a model based on several independent non-differentiable loss functions simultaneously; for example, the F1-score, which is used during testing, can be used during training when a gradient-free optimization algorithm is utilized. Furthermore, the training (i.e., optimization of weights) in any DNN can be possible with a small size of the training dataset. To address these concerns, we propose an efficient version of the gradient-free Coordinate Search (CS) algorithm, an instance of General Pattern Search (GPS) methods, for training (i.e., optimizing) neural networks. The proposed algorithm can be used with non-differentiable activation functions and tailored to multi-objective/multi-loss problems. Finding the optimal values for weights of ANNs is a large-scale optimization problem. Therefore instead of finding the optimal value for each variable, which is the common technique in classical CS, we accelerate optimization and convergence by bundling the variables (i.e., weights). In fact, this strategy is a form of dimension reduction for optimization problems. Based on the experimental results, the proposed method is comparable with the SGD algorithm, and in some cases, it outperforms the gradient-based approach. Particularly, in situations with insufficient labeled training data, the proposed CS method performs better. The performance plots demonstrate a high convergence rate, highlighting the capability of our suggested method to find a reasonable solution with fewer function calls. As of now, the only practical and efficient way of training ANNs with hundreds of thousands of weights is gradient-based algorithms such as SGD or Adam. In this paper we introduce an alternative method for training ANN.

Keywords: Automated Algorithm, Bio-inspired, Decision Making
Abstract: In evolutionary algorithms and metaheuristics, defining when applying a specific operator is important. Besides, in complex optimization problems, multiple populations can be used to explore the search space simultaneously. However, one of the main problems is extracting information from the populations and using it to evolve the solutions. This article presents the inequality-based multi-population differential evolution (IMDE). This algorithm uses the K-means to generate subpopulations (settlements). Two variables are extracted from the settlements, the diversity and the Gini index, which measure the solutions’ distribution and the solutions’ inequality regarding fitness. The Gini index and the diversity are used in the IMDE to dynamically modify the scalation factor and the crossover rate. Experiments over a set of benchmark functions with different degrees of complexity validate the performance of the IMDE. Besides comparisons, statistical and ranking average validate the search capabilities of the IMDE.

Keywords: Defense and Security, Deep Learning
Abstract: Recently, deep neural networks (DNNs) have been widely used and proven successful in many real-world tasks. There are many third-party DNN services available for data holders who want to develop custom DNN applications for their data and tasks. To ensure data privacy, it is crucial to safeguard the data holder's training data. This paper explores a unique attack paradigm where a hostile third-party DNN model supplier subtly obtains training data from the data holder. Prior attacks which can steal training data typically use augmented datasets to memorize the information of the data that the attacker intends to steal. However, these attacks are easily identified since the augmented datasets are visually different from the original dataset and rendered ineffective. In this attack, we generate an augmented dataset by modifying a portion of the training data using the DNN-based image steganography technique. This approach creates an augmented dataset that is visually identical to the original training dataset, making it difficult for humans to detect. Through extensive experiments, we have successfully and quietly accessed the confidential training data of data holders.

Keywords: E-health, Deep Learning, Decision Making
Abstract: In healthcare, building a supervised learning system faces the challenge of access to a large, labeled dataset. To overcome this problem, we propose a deep transfer learning method that addresses imbalanced data problems in healthcare, focusing on structured data. We use publicly available breast cancer datasets to generate a source model and transfer learned concepts to predict high-grade malignant tumors in patients diagnosed with breast cancer at Mayo Clinic. The diabetes dataset is then used to generalize the transfer learning idea. We compare our results with state-of-the-art techniques and demonstrate the superiority of our proposed methods. Our experiments on breast cancer data under simulated class imbalanced settings further demonstrate the proposed method's ability to handle different degrees of class imbalance. We conclude that deep transfer learning on structured data can efficiently address imbalanced class and poor performance learning on small dataset problems in clinical research.

Keywords: E-health, Deep Learning, Explainability
Abstract: In medical applications, artificial intelligence (AI) methods have achieved considerable progress in various areas and also in newborn screening programs. In particular, interpretable AI methods have been applied in newborn screening aiming to increase analytical specificity and predictive power of screening results. In this study, we apply ensemble and deep learning methods in newborn screening for isovaleric aciduria (IVA) on a data set containing more than 2 million newborns. We show that these methods can reduce the number of newborns falsely classified with IVA by 100% with Extreme Gradient Boosting (XGBoost), by 78.94% with Random Forest (RF), and by 78.94% with Feed Forward Neural Networks (FFNN) compared to currently applied newborn screening methods. Furthermore, we show how explainable AI (XAI) methods can be used to interpret these black-box classification results and further apply them for potential biomarker discovery. The XAI methods reveal that besides the biomarker isovaleryl carnitine (C5), the birth year and the amino acid tryptophan (Trp) are influential in reducing the false positive rate. By this, we show that ensemble and deep learning could be highly beneficial in newborn screening and could have a major impact on newborns and their families, as it reduces false positive screening results and guides new directions for future research in this field.

Keywords: Image Processing, E-health, Ensemble Learning
Abstract: Oral squamous cell carcinoma (OSCC) is a significant health issue in the oral cancer domain; a screening tool for timely and accurate diagnosis is essential for effective treatment planning and prognosis in patients' life expectancy. In this paper, we address the problem of object detection and classification in the context of OSCC, by presenting a comparative analysis of three state-of-the-art architecture: YOLO, FasterRCNN, and DETR. We propose a deep learning ensemble model to address both object detection and classification problem leveraging the strengths of individual models to achieve higher performance than single models. The proposed architecture was evaluated on a real-world dataset developed by experienced clinicians who manually labeled individual photographic images, producing a benchmark dataset. Results from our comparative analysis demonstrates the ensemble detection model achieves superior performance compared to the individual models, outperforming the average value of the individual models' map@50 metric by 24% and the value of the map@95-50 metric by 44%

Keywords: Ensemble Learning, Data Mining, Model-Based
Abstract: A number of the genetic network inference methods have been proposed. These methods have been basically designed to analyze gene expression data of bulk cells. Recently, on the other hand, researchers have been capable of using gene expression data measured at single-cell resolution. The existing inference methods are however incapable of analyzing time-series of single-cell data because of high cell-to-cell variation in gene expression. This study therefore proposed the new inference method that has an ability to analyze steady-state and pseudo time-series of single-cell gene expression data. The pseudo time-series data are obtained through the pseudo-temporal ordering analysis. As the precise information about the measurement time is unavailable in pseudo time-series data, our method infers a genetic network using the signs of time derivatives of gene expression levels, that can be estimated from the given data. Through the numerical experiments, we finally confirmed the effectiveness of the proposed method.

Keywords: Deep Learning, Explainability, E-health
Abstract: With application contexts ranging from psychophysiology to neuromarketing, electroencephalography (EEG)-based emotion recognition is a fundamental technology for affective computing. In this context, EEG signals can be processed via artificial neural networks (NNs) to achieve accurate recognition of users' emotions. Still, NNs are rarely employed in real-world decision-making processes, since their internal model works as a hardly trustable black box. A NN's reasoning can be explained in a human-comprehensible manner by exploring its latent space to understand if some domain knowledge is actually represented and exploited for the classification. Those approaches assume that a trained NN autonomously organizes its latent space according to some domain concepts to process the data via human-like reasoning. However, there is no guarantee that such an assumption holds, since the latent space is not built for this aim. On the other hand, forcing the organization of the latent space (e.g. via contrastive learning) can result in poor recognition performances due to information loss. To guarantee great recognition performances and provide a domain-knowledge-driven organization of NNs' latent space, we combine the well-known training procedure based on a categorical cross-entropy loss with a supervised contrastive learning approach for continuous values labels. The proposed approach (i) enables the explanation of NN's reasoning in terms of the importance of high-level domain concepts in the final classification, and (ii) results in a recognition performance comparable to or better than the one achieved via an approach based solely on maximizing recognition. The proposed approach is tested on the publicly available MAHNOB dataset.

Keywords: E-health, Evolving Learning, Model-Based
Abstract: The inverse problem in electrocardiography is an ill-posed problem where the objective is to reconstruct the electrical activity of the epicardial surface of the heart, given the electrical activity on the thorax’ surface. In the forward problem, the electrical propagation from heart to thorax is modeled by the volume conductor equation with Dirichlet boundary conditions in the heart’s surface, and null flux coming from the thorax. The inverse problem, however, does not have a unique solution. In order to find solutions for the inverse problem, techniques such as Tikhonov regularization are classically used, but they often deliver unrealistic solutions. As an alternative, we propose a novel approach, where a fixed solution of the volume conductor model with a source in a forward scheme is used to solve the inverse problem. The unknown values for parameters of the fixed solution can be found using optimization techniques. Due to the characteristics of the problem, where each single evaluation of the cost function is expensive, we use a specialized CMA-ES-based Bayesian optimization technique, that can deliver good results even with a reduced number of function evaluations. Experiments show that the proposed approach can deliver improved results for in-silico simulations.

Keywords: Robotics, Deep Learning, Autonomous Systems
Abstract: Recent studies have demonstrated that active perception can improve the perception abilities of deep learning (DL) models. However, there are challenges associated with using active perception in DL models, including the need for datasets and/or realistic simulations that can support the training process, along with the difficulty of predicting the final target position, which reduces planning efficiency. To address these challenges, this work presents a methodology for enhancing the perception abilities of DL models through active perception. The methodology proposes a way to create datasets for active perception by fusing existing large-scale datasets and decomposing the active perception problem into three sub-tasks for face recognition. The sub-tasks aim to determine the appropriateness of the current view for face recognition, the direction in which the robot should move for a better viewpoint, and the expected amount of movement required. A novel trial-based approach is introduced to estimate the final target position, making the method platform-agnostic and easily applicable to different robots. The proposed methodology is validated through experiments on two well-known face verification datasets that have been augmented with occlusions, demonstrating its effectiveness in enhancing the perception abilities of DL models through active perception.

Keywords: Federated Learning, Deep Learning
Abstract: In federated learning scenarios, data heterogeneity can significantly impact performance. Personalized federated learning seeks to provide individualized models for each client to enhance convergence on heterogeneous data. We discovered that initially training the personalized layers, also known as the head, of the model first can alleviate the effects of data heterogeneity. As a result, we propose a simple method named FedLoop. This method uses a loop topology structure, eliminating the need for a central server or data exchanges between participants, thereby safeguarding privacy. Within FedLoop, clients act as nodes in a loop. The training process for each node consists of two phases: an initial phase solely for the personalized layers and a subsequent phase dedicated to the training of all layers. This looping process continues until a set round limit is achieved. Experimental findings reveal that FedLoop outperforms the existing state-of-the-art algorithm, FedALA. FedLoop effectively addresses challenges posed by data heterogeneity and its rapid convergence significantly cuts down communication overheads in federated learning.

Keywords: Intelligent Control, Bio-inspired
Abstract: Abstract—This paper presents the design of an intelligent controller for uncertain discrete-time nonlinear systems. The proposed controller is resilient to external unknown disturbances as well as state and input uncertainties, even though the model of the system is considered unknown. An intelligent controller is designed for an unknown discrete-time nonlinear system, this controller is model-free and it is based on sensor measurements and therefore including unknown system dynamics, actuator nonlinearities, measurement errors, noise, uncertainties, external disturbances and other phenomena associated to real-world applications. Then, using the neural model, a backstepping controller is designed to ensure a resilient performance. Finally, real-time results are included to demonstrate the effectiveness of the proposed approach using a three-phase induction motor. Index Terms—Uncertain discrete-time nonlinear system, Intelligent control, Neural control, Resilience, Experimental results, Induction motor

Keywords: Fuzzy Systems, Transportation and Vehicle Systems, Decision Making
Abstract: Intelligent Transportation Systems rely heavily on the Vehicular Ad-hoc Network to enhance road safety and comfort. This research proposes and evaluates an intelligent backoff management scheme, utilizing Adaptive Neuro-Fuzzy Inference System (ANFIS), for the Vehicular Ad-hoc Networks. The proposed scheme is trained by TensorFlow to adjust the contention window size at the MAC layer of IEEE 802.11p. Taking into account the local density, local spatial distribution, and successful/unsuccessful transmission records, each transmitting node can determine the best contention window value for transmitting packets. This scheme effectively mitigates packet collisions, ensuring a high packet delivery ratio and average throughput, along with a low average end-to-end delay for various network scenarios. Simulation results confirm the efficiency of the proposed scheme and also show that it outperforms the conventional IEEE 802.11p method and other recent protocols.

Keywords: Deep Learning, Autonomous Systems, Fault Detection
Abstract: Anomalies in the domain of autonomous driving are a major hindrance to the large-scale deployment of autonomous vehicles. In this work, we focus on high-resolution camera data from urban scenes that include anomalies of various types and sizes. Based on a Variational Autoencoder, we condition its latent space to classify samples as either normal data or anomalies. In order to emphasize especially small anomalies, we perform experiments where we provide the VAE with a discrepancy map as an additional input, evaluating its impact on the detection performance. Our method separates normal data and anomalies into isolated clusters while still reconstructing high-quality images, leading to meaningful latent representations.

Keywords: Data Mining, Deep Learning, Pattern Recognition
Abstract: Lexicons are a fundamental resource for sentiment analysis, offensive language identification, trend detection, and document classification. Lexicons have the advantage of being easy to use, but most of the existing lexicons have been created manually. Recently, researchers have been interested in extending the use of lexicons to different fields. In this paper, an easy-to- compute statistics-based method for extracting lexicons in specific domains or ad-hoc lexicons is shown. The proposed method was evaluated on two datasets and achieved 80% accuracy in document classification. This novel approach is expected to be a valuable tool for researchers and practitioners who need to quickly and efficiently create domain-specific and ad-hoc lexicons

Keywords: Automated Algorithm, Decision Making, Federated Learning
Abstract: Author profiling is the process of analysing text to determine one or more identifying characteristics of the author, mostly used to determine key demographic information. This type of classification problem is ideally suited to machine learning approaches. In this study, a new transfer learning approach is introduced using a pre-trained XLNet language model which was then fine-tuned to the specific author profiling task. Informed by previous literature, a Support Vector Machine, Feed-Forward Neural Network, and Convolution Neural Network were also developed for comparison. These algorithms were used to predict gender and age group on a single training and testing domain. As a model that works across multiple domains is desirable, each model was also tested on two domains which were independent of the training domain. The results demonstrated that the transfer learning model is superior to the other methods used for comparison in this study. Although applying the transfer learning model to the cross-domain context decreased its performance, it was still able to achieve a higher degree of accuracy on one testing domain than the Support Vector Machine which was trained and tested on that same domain. In addition, some interesting results emerged regarding the transfer of hyperparameter performance between tasks that share a common factor, be that classification task or training domain.

Keywords: Autonomous Systems, Deep Learning, Defense and Security
Abstract: This paper considers the adaptation of the e-coaching concept at times of emergencies and disasters, through aiding the e-coaching with intelligent tools for monitoring humans' affective state. The states such as anxiety, panic, avoidance, and stress, if properly detected, can be mitigated using the e-coaching tactic and strategy. In this work, we focus on a stress monitoring assistant tool developed on machine learning techniques. We provide the results of an experimental study using the proposed method.

Keywords: Deep Learning, Decision Making, Autonomous Systems
Abstract: The impacts of mass migration, such as crises induced by climate change, extend beyond environmental concerns and can greatly affect social infrastructure and public services, such as education, healthcare, and security. These crises exacerbate certain elements like cultural barriers and discrimination by amplifying the challenges faced by these affected communities. This paper proposes an innovative approach to address migration crises in the context of crisis management through a combination of modeling and imbalance assessment tools. By employing deep learning for forecasting and integrating causal reasoning via Bayesian networks, this methodology enables the evaluation of imbalances and risks in the socio-technological landscape, providing crucial insights for informed decision-making. Through this framework, critical systems can be analyzed to understand how fluctuations in migration levels may impact them, facilitating effective crisis governance strategies.

Keywords: Decision Making, Biometric Systems, Pattern Recognition
Abstract: This study offers a causal probabilistic modeling for inferring the relationship between humans' cognitive load, the physiological signal predictors of such load, and personality traits. We selected a subset of such signals (heart rate, intervals between successive heartbeats, galvanic skin response, and temperature) from the CogLoad dataset using wearable devices. Structural Equation Modeling techniques were employed to select the predictors to identify the level of cognitive load, for which the ground truth was assessed using subjective tests such as HEXACO that determine the personality traits of the human subjects. Bayesian networks were deployed to investigate the causal relationship and model the inference scenarios. The proposed model is intended to contribute to developing a Computational Intelligence tool for monitoring social health in scenarios of future potential crises such as pandemics and mass migration.

Keywords: Decision Making, Pattern Recognition, E-health
Abstract: Exploration and analysis of changes in human biometrics, such as heart rate and blood pressure associated with exposure to traumatic events is the primary goal of this article. We aimed at answering the questions on whether there is a significant difference in biometrics observed in the peaceful and disaster times. Overall, we developed and tested a new technique to measure the difference in the indicators of stress during relatively peaceful times, and during natural and human-made disasters and crises. The proposed approach holds significant potential in the context of e-health and mass migration, offering a valuable tool to recognize and address stress in traumatic events resulting from, for example, forced displacement, armed conflicts, and the impacts of climate change.

Keywords: Deep Learning, Image Processing, Pattern Recognition
Abstract: This paper offers an investigation into leveraging computational intelligence (CI) for the assessment of stroke-induced motor weakness in post-stroke survivors, serving as an indicator of neurological function. The proposed methodology deploys deep learning algorithms to analyze video recordings obtained during the post-stroke hospitalization phase. The model effectively categorizes the degree of stroke-induced weakness in the patient's left arm across two and three distinct classes aligned with the National Institutes of Health Stroke Scale. This study was motivated by the limitations of existing monitoring technologies, such as using pressure sensing mattresses (such as low resolution and low accuracy). Our long-term strategy is to deploy several means for monitoring the patients' motor function. This study demonstrated a binary classification model using video data collected from a cohort of 23 post-stroke patients in a clinical setting for 48 hours. Employing a 3-fold cross-validation methodology, the developed model showcases an accuracy rate of 92.10 ± 4.08% for the binary classification, distinguishing between mild and severe stroke-induced weakness in the left arm. In the case of three classes, the model achieves an accuracy of 89 ± 4.95%.

Keywords: Multi-Agent System, Robotics
Abstract: The complexity of multi-robot systems (MRS) involves the challenging task of robot coordination to achieve the system’s goal. That indicates the necessity to integrate automated planning to mitigate disruptions and continually adjust the behavior of robots in the presence of failures. Adequate architectures to integrate MRS with automated planning present a gap in the literature, indicating the necessity for further research. To address this gap, we present the Multi-Robot System Architecture with Plan (MuRoSA-Plan) for mission coordination of heterogeneous robots illustrated with a healthcare service case. This work contribution is the MuRoSA-Plan architecture to MRS domain applications focusing on plan recovery. The experimental results show that MuRoSA-Plan generates runtime-adapted plans satisfying the goals of the multi-robot coordination case mitigating mission disruptions.

Keywords: Reinforcement Learning, Intelligent Control, Human-Computer Interactions
Abstract: Offline reinforcement learning provides a viable approach to obtain advanced control strategies for dynamical systems, in particular when direct interaction with the environment is not available. In this paper, we introduce a conceptual extension for model-based policy search methods, called variable objective policy (VOP). With this approach, policies are trained to generalize efficiently over a variety of objectives, which parameterize the reward function. We demonstrate that by altering the objectives passed as input to the policy, users gain the freedom to adjust its behavior or re-balance optimization targets at runtime, without need for collecting additional observation batches or re-training.

Keywords: Multi-Agent System, Advanced Optimization, Ethical AI
Abstract: The concept of stable matching is substantially used in bipartite graphs with individual preferences of the vertices. The existence of stability restricts the weight and size of the matching to be satisfactory. We study the trade-offs in stability, weight and cardinality in a one-to-many capacitated weighted bipartite matching with an edge-weight-oriented preference setting. We establish a stability relaxation framework which is adaptive to the pairing suitability and capacity of the vertices. The purpose of the relaxation is to update the stable matching towards the balance of stability, weight and cardinality in the result. The relaxation preserves fairness by keeping the satisfaction degradation of the vertices with the potential new partner in a desired range. We propose an algorithm to produce a new matching using the stability relaxation framework. Furthermore, we define a novel popularity measurement model of matching based on the edge weight with the multi-voting ability of one-sided vertices. We show the resulting matching is also popular as stable matching. The experimentation performed based on the use case of the homeless placement system complements the claim of improving the weight and cardinality in the matching with marginal and fair relaxation of stability.

Keywords: Agent-Based Modeling, Multi-Agent System, Reinforcement Learning
Abstract: Cooperation is a defining trait of Multi-Agent Systems. At the centre of these systems lies a communication network which governs how information flows from one agent to the next. However, the design of these networks is often overlooked despite the profound impact it can have on both the task performance of the agents and the emergent phenomena they produce. In this work we aim to illustrate this by investigating whether network centrality impacts the task performance and emergent inequality (unequal distribution of resources) of resource gathering agents.

We achieve this by constructing several communication networks with increasing centrality and use them with an Agent-Based Model called GATHER. Our results indicate that as the variance of the population's centrality increases, the task performance of an agent population will decrease. Furthermore, we demonstrate that simply changing the centrality of the network can produce distinct results and emergent phenomena (inequality or the lack thereof in our case). We then further support this claim by increasing the reciprocity of one of our communication networks which results in a system with greater task performance and significantly lower inequality, further illustrating the impact communication network topology can have on Multi-Agent Systems.

Keywords: Multi-Agent System, Autonomous Systems
Abstract: Multi-agent systems can be prone to failures during the execution of a mission, depending on different circumstances, such as the harshness of the environment they are deployed in. As a result, initially devised plans for completing a mission may no longer be feasible, and a re-planning process needs to take place to re-allocate any pending tasks. There are two main approaches to solve the re-planning problem (i) global re-planning techniques using a centralized planner that will redo the task allocation with the updated world state and (ii) decentralized approaches that will focus on the local plan reparation, i.e., the re-allocation of those tasks initially assigned to the failed robots, better suited to a dynamic environment and less computationally expensive. In this paper, we propose a hybrid approach, named GLocal, that combines both strategies to exploit the benefits of both, while limiting their respective drawbacks. GLocal was compared to a planner-only, and an agent-only approach, under different conditions. We show that GLocal produces shorter mission make- spans as the number of tasks and failed agents increases, while also balancing the trade-off between the number of messages exchanged and the number of requests to the planner.

Keywords: Advanced Optimization, Deep Learning
Abstract: The current advances in generative artificial intelligence for learning large neural network models with the capability to produce essays, images, music and even 3D assets from text prompts create opportunities for a manifold of disciplines. In the present paper, we study the potential of deep text-to-3D models in the engineering domain and focus on the chances and challenges when integrating and interacting with 3D assets in computational simulation-based design optimization. In contrast to traditional design optimization of 3D geometries that often searches for the optimum designs using numerical representations, e.g. B-Spline surfaces, natural language challenges the optimization framework by requiring a different interpretation of variation operators while at the same time may ease and motivate the human user interaction. Here, we propose and realize a fully automated evolutionary design optimization framework using Shap-E, a recently published text-to-3D asset network by OpenAI, in the context of aerodynamic vehicle optimization. For representing text prompts in the evolutionary optimization, we evaluate (a) a bag-of-words approach based on prompt templates and Wordnet samples, and (b) a tokenisation approach based on prompt templates and the byte pair encoding method from GPT4. In our experiments, we show the text-based representations allow the optimizer to find better performing designs. However, it is important to ensure that the designs generated from prompts are within the object class of application, i.e. diverse and novel designs need to be realistic. Furthermore, more research is required to develop methods where the strength of text prompt variations and the resulting variations of the 3D designs share causal relations to some degree to improve the optimization.

Keywords: Fault Detection, Financial Engineering, Deep Learning
Abstract: Abstract—False data injection detection is a topic of interest because systems are prone to cyberattacks which can manipulate the state estimation process by injecting malicious data into the measurements, bypassing the detection of the security system. Causing the results of the state estimation to deviate from the safe values. This work proposes a false data injection detection methodology based on deep neural networks using sliding windows to generate online error vectors in order to detect and classify malicious data from measurement data. Two multilayer perceptron deep neural networks and the convolutional neural network were used in this work. In order to verify the feasibility of the proposed methodology, it is tested on data daily closing prices of the S&P 500 Index, pulled from Yahoo Finance for the years 2013–2022 to which false data were injected via software. The results show that the convolutional neural network presents the best results, with an accuracy above 93% and an F1-score of 0.91. It is shown that deep neural networks are a powerful tool in the detection of false data in data obtained through measurements.

Keywords: Fault Detection, Internet of Things, Signal Processing
Abstract: In today’s world, sensors play a crucial role, as they feed information to make accurate decisions and take actions; therefore, making sure that sensors behave correctly is critical. In this work, we focus on inspecting the data provided by sensors, aiming at discovering any issue due to malfunction, misuse, or any other source of error before the issue is propagated through the system. To achieve that, we propose a novel approach based on wavelets embedded in a microcontroller to analyze data from sensors. The objective is to determine whether the sensor is issuing correct data (normal behavior) or not (abnormal behavior), to prevent the error from reaching other parts of the system.

Keywords: Deep Learning, Fault Detection, Internet of Things
Abstract: A complex system that is governed by several smaller sub-systems whose coordinated functionality allows it to work properly over time can be challenging to analyze for faults on real time by an observer; moreover, if such failing system could work with no obvious signs of fault over time until it becomes catastrophic and clearly identifiable. Because the variables involved in such system's functionality are usually not easily correlated, the different time-series they might generate can be extremely difficult to analyze by conventional means. Lately, 2-dimensional Convoluted Neural Networks (2D-CNN) have been used to introduce artificial intelligence into diagnosis and fault detection with success; however, the systems that so far have benefited from this are mainly those that deal with images, like medical diagnosis using x-ray images, or autonomous driving using real time pictures, although recently, some resent research on robotic sensor fault and signal analysis have been published using 1-dimensional CNN (1D-CNN) for time-domain signals. This paper proposes a novel 2D-CNN approach to fault identification of an unknown, discrete-time, non-linear system; by recognizing features that are consistent with a fault in a signal-image of several layers. With such signal-image being an artificial picture created by combining all the system signals in a single high-layered image format that is recognizable by a conventional 2D-CNN. This paper also includes the results of its applicability in a fault identification of a three-phase induction motor in a simulation environment and with measurements of a real motor with injected faults.

Keywords—deep learning, fault identification, fault prediction, applied artificial intelligence. Convolutional neural networks non-linear systems.

Keywords: Fault Detection, Model-Based, Deep Learning
Abstract: The utilization of computational intelligence, particularly Artificial Neural Networks (ANNs), for fault detection is of paramount importance as it empowers industries to proactively identify anomalies, leading to improved system reliability, reduced downtime, and enhanced safety. By leveraging the pattern recognition capabilities of ANNs, complex data patterns indicative of faults can be accurately identified and analyzed in real-time, enabling early intervention and preventing potential catastrophic failures. Additionally, the importance of fault detection in refrigeration systems lies in its ability to proactively identify and address potential issues, ensuring optimal performance, energy efficiency, and longevity of the system while preventing costly breakdowns and ensuring product safety and quality. The main aim of this study is to create a computational intelligence model that can accurately depict the energy and exergy performance of a GAX hybrid refrigeration system. Moreover, the model aims to identify potential instrument failures occurring at different parts of the system. The primary findings indicate that creating a numerical database using the governing equations of the GAX system enables the identification of anomalies in the instrumental measurements of operating parameters. Subsequent research aims to incorporate experimental data from a broader range of parameters, encompassing additional sections of the GAX system.

Keywords: Fault Detection, Autonomous Systems, Image Processing
Abstract: Despite the significant achievements in the development of automation technologies, the application of autonomous robots to improve the production efficiency of small-scale industries has been largely ignored. While there has been excellent progress in industrial image processing systems implementation, most of the work has focused on a unique aspect of specific objects rather than introducing a general inspection system. Thus, this paper discusses the critical industrial topic of quality control, which develops rapidly through the use of autonomous systems. Given the high cost of implementing automated systems, this paper presents an affordable low-budget solution for the visual inspection system. This method of inspecting screw dimensions consists of four visual inspection parts and a special mechanical supporting structure. The designed system was able to check the overall screw dimensions, including screw head diameter, screw head driven type, screw length, screw thread length, and screw head thickness. It could also separate the qualified screws from the unqualified ones after the inspection process. The accuracy of most inspection cases is 100%, meaning the error ranges within 0.1mm, which meets all the non-negotiable requirements and most of the target requirements. The visual inspection parts can be further enhanced by building a template matching library that includes different angles of the screw head or by using Hough Transform to identify the defect types of the screw thread.

Keywords: Transportation and Vehicle Systems, Operations Research, Model-Based
Abstract: In sea-road intermodal container terminals, the integrated scheduling problem for the multi-stage transshipment system (ISP_MST_CT) is influenced by factors such as the number of containers, multi-stage interactions, and various types of equipment, making it challenging to construct the model. Additionally, assigning an appropriate number of AGVs to the transshipment tasks can significantly avoid resource waste at the terminals. This paper, for the first time, considers the ISP_MST_CT of quay cranes, AGVs, yard cranes, and external trucks, encompassing four operational stages. A mixed-integer programming model is formulated to simultaneously optimize the maximum completion time, total energy consumption of quay cranes and yard cranes, and total waiting time of AGVs. The nondominated sorting genetic algorithm II (NSGAII) algorithm is employed to solve this problem. The experiment results validate that NSGAII is capable of efficiently solving ISP_MST_CT of different scales and obtaining superior solutions within a short time. Furthermore, a series of experiments with 20 containers demonstrates that 8 AGVs can keep the balance among the three optimization objectives, while reducing the waste of AGVs and providing valuable insights to terminal managers.

Keywords: Operations Research, Data Mining, Transportation and Vehicle Systems
Abstract: For the truck appointment system in a container terminal, optimizing the configuration of gate lane and yard crane based on the appointment information is the key to shorten the external truck waiting time and reduce the redundancy of terminal resource. A hybrid approach combining deep neural network and optimization model is proposed. The deep neural network is applied to predict the truck waiting time in the yard based on the yard data. The optimization configuration model for gate lane and yard crane is established by combining the predicted result. The average waiting time of trucks, the configuration of gate lanes and yard cranes before and after optimization are compared. The results show the effectiveness of the proposed approach, which also provides a new road map for optimizing container terminal resource configuration.

Keywords: Operations Research, Transportation and Vehicle Systems
Abstract: Dynamic Berth Allocation Problem (DBAP) is an essential problem in container terminal operations. Most studies focus on discrete or continuous berths in DBAP. However, affected by the geographical conditions, the mixture of discrete and continuous berths which are called hybrid berths often appear in real port container terminals. Moreover, the arrival time of vessels is often fluctuant due to the influence of environmental factors. To solve such a Dynamic Hybrid Berth Allocation Problem (DHBAP) under vessels’ arrival delay, this study develops a proactive-reactive approach. Specifically, we establish a mixed-integer programming model with a buffer as the proactive strategy to obtain a baseline schedule. Then, we propose a hybrid berth reactive strategy (HBRS) to adjust the baseline schedule for vessels that are delayed. To get a better solution in a short time, a genetic algorithm is designed. We verify the effectiveness of the proposed HBRS by comparing it with the most commonly used right-shift strategy. Experimental results show that the longer the buffer is, the better the robustness of the model is, but the total time of the vessel in terminals will also increase. Compared with the right-shift strategy, the proposed HBRS can obtain an allocation plan with similar robustness in a shorter total time of the vessel in terminals.

Keywords: Decision Making, Advanced Optimization, Electric Vehicle
Abstract: There is urgency for electrifying fleet vehicles as a means to reach net-zero emissions and promote sustainability, including at maritime container ports. Ports are exploring the incorporation of electric terminal tractors and supporting infrastructure in an effort to minimize the environmental effects of their operations while simultaneously improving service performance. The challenges include planning of investments in infrastructure that will meet charging requirements of these terminal tractors while maintaining operational efficiencies. This paper develops an optimization and associated risk register for strategic capacity expansion of electric vehicle fleets at maritime container ports. The approach includes multi-criteria decision analysis (MCDA) and a characterization of enterprise risk as a disruption of system order. A demonstration of schedule optimization uses linear programming models for thirty-two combinations of plug-in, wireless, and wireless dynamic charging infrastructure configurations to determine optimal charger locations. In a robust ensemble model, the optimization accompanies a comprehensive risk analysis that disrupts importance orders across seven scenarios: (1) Environmental Change, (2) Policy Revision, (3) Technology Innovation, (4) Cyber Attack, (5) Market Shift, (6) Electrical Grid Stress, and (7) Workforce Interruption. The results support the decisions and enterprise risk management for a 1.5 billion strategic plan for port infrastructure. The plan involves selecting charging station locations, determining charging schedules, and selecting charger models while considering multiple performance criteria such as safety, operational efficiency, cost-effectiveness, and reliability. The approach is generally applicable for a variety of complex systems to mitigate schedule and cost risks while improving sustainability. The audience of the paper includes owners and operators of transportation and energy infrastructures, asset managers, logistics service providers, and others.

Keywords: Operations Research, Decision Making
Abstract: In this paper, we propose to design a large-scale intelligent collaborative platform for freight forwarders. This platform is based on a mathematical programming formulation and an efficient solution approach. Forwarders are middlemen who procure container capacities from carriers and sell them to shippers to serve their transport requests. However, due to demand uncertainty, they often either over-procure or under-procure capacities. We address this with our proposed platform where forwarders can collaborate and share capacities, allowing one’s transport requests to be potentially shipped on another forwarder’s container. The result is lower total costs for all participating forwarders. The collaboration can be formulated as an integer linear program we call the Freight Forwarders’ Collaboration Problem (FFCP). It is a variant of the bin-packing problem, hence it is NP-Hard. In order to solve large-scale FFCP instances efficiently, we propose a two-step approach involving an initial greedy assignment followed by a fine-tuning step. Computational experiments have shown that our approach can offer a significant reduction of run-time between 77% and 97%, without any loss of solution quality.

Keywords: Particle Swarm Optimization, Ensemble Learning, Swarm Intelligence
Abstract: As a subset of metaheuristics, nature-inspired optimization algorithms such as particle swarm optimization (PSO) have shown promise both in solving intractable problems, and in their extensibility to novel problem formulations due to their general approach requiring few assumptions. Unfortunately, a given algorithm requires detailed tuning of parameters and cannot be proven to be best suited to a particular problem class on account of the “no free lunch” (NFL) theorems. Using these algorithms in real-world problems requires exquisite knowledge of the many approaches and applying them based upon intuition. This research aims to present a unified view of PSO-based approaches from the perspective of relevant systems engineering problems, with the purpose to then elicit the best solution for any problem formulation in an ensemble learning approach. The central hypothesis of the research is that using the PSO algorithms found in literature to solve real-world optimization problems requires a general ensemble-based method for all problem formulations but a single implementation and solution for any instance. The main results will be a problem-based literature survey and a general method to find more globally optimal solutions for any systems engineering optimization problem.

Keywords: Ethical AI, Human-Computer Interactions, Robotics
Abstract: As Industry 5.0 emerges by blending advanced technologies with human-centered approaches, the integration of machine learning (ML) in human-robot collaboration (HRC) becomes increasingly prominent. This paper explores the philosophy and ethics underlying the application of machine learning in Industry 5.0, specifically focusing on HRC. It examines the ethical considerations, philosophical implications, and potential challenges that arise in this evolving paradigm. The paper emphasises the need for a thoughtful and ethical approach to ensure the beneficial and responsible use of ML in Industry 5.0.

Keywords: Ethical AI
Abstract: Building responsible and trustworthy AI solutions is now the norm, yet the challenge of bridging the ethical AI principles to practice gap especially for small to medium businesses is significant with the forthcoming European Union AI ACT (2023) becoming a major disrupter for global businesses. AI Adoption by SMEs is growing but there are many barriers including limited AI skills, complexity of projects, lack of understanding of what is trustworthy and responsible AI and the tools needed to consequence scan the wider impact on stakeholders of innovative products and services prior to market. This paper presents a case study of the Greater Manchester AI Foundry (GMAIF) - a consortium model for University – SME collaboration designed to foster ethical and responsible AI design and development practices into SMEs and new start-ups. The GMAIF model supports the creation of proof-of-concept demonstrator projects, forming a number of tangible products or eservices, that demonstrate how AI can be used to enhance or provide new products and services. Whilst the model is demonstrated within the UK, its concepts are generalizable and applicable globally. GMAIF has impacted 186 SMEs in the UK, with 67 new AI products being developed by SMEs and an additional 80 innovative products to market.

Keywords—ethics artificial intelligence, toolkits, responsible technology, industry, SME

Keywords: Ethical AI, Robotics, Human-Computer Interactions
Abstract: In the era of Industry 5.0, the convergence of humans and robots in collaborative work environments has brought forth the concept of digital twins (DTs) of humans and robots. These virtual replicas, mirroring their physical counterparts, have become integral to the design and operation of complex systems. This paper aims to explore the ethical and philosophical implications associated with the design and use of DTs of humans and robots in human-robot collaboration (HRC), and even further in human-robot teaming (HRT). By examining the potential benefits, challenges, and risks, this research seeks to shed light on the responsible development and application of DTs in the context of Industry 5.0.

Keywords: Human-Computer Interactions, Robotics, Ethical AI
Abstract: This paper delves into the ethical, philosophical, and practical dimensions associated with the transition from caged robots to human-robot teaming (HRT). By exploring the evolving dynamics between humans and robots, this paper examines the ethical challenges, philosophical implications, and practical considerations that arise as collaboration and integration between humans and robots deepen. It emphasises the need for responsible design, implementation, and ethical frameworks to guide the development and deployment of human-robot teams. Particular focus is put into the ethical ramifications of choosing between rigid and soft actuators. The study underscores the significance of employing admittance and impedance control techniques to regulate interaction forces and compliance between humans and robots. By analysing the ethical implications of utilising soft actuators, the paper emphasises the potential advantages, such as enhanced safety and reduced risk of harm during close human- robot collaboration.

Keywords: Ethical AI, Decision Making, E-health
Abstract: AI technologies have made significant advancements across various sectors, especially healthcare. Although AI algorithms in healthcare showcase remarkable predictive capabilities, apprehensions have emerged owing to errors, biases, and a lack of transparency. These concerns have led to a decline in trust among clinicians and patients, while also posing the risk of further accentuating pre-existing biases against marginalized groups and exacerbating inequities. This paper presents a scenario-based preferences risk register framework for identifying and accounting AI algorithm biases in diagnosing diseases. The framework is demonstrated with a realistic case study on cardiac sarcoidosis. The framework identifies success criteria, initiatives, emergent conditions and the most and least disruptive scenarios. The success criteria align with the National Institute of Standards and Technology AI Risk Management Framework (NIST AI RMF) trustworthy AI characteristics, and the scenarios are based on various statistical/computational bias that causes algorithmic bias. The framework provides valuable guidance for leveraging AI in healthcare, enhancing objective designs, and mitigating risks by adopting a figure of merit to score the initiatives and measuring the disruptive order. By prioritizing transparency, trustworthy AI, and identifying the most and least disruptive scenarios/biases, the framework promotes responsible and effective use of AI technologies in healthcare.

Keywords: Explainability, Agent-Based Modeling
Abstract: Machine learning models can make accurate predictions but trust in the models depends on being able to understand why those predictions were made. Unfortunately, machine learning models are black boxes making interpretation difficult. Previously we used an evolutionary algorithm to evolve triplets of neural network players for instances of the Generalized Divide-the-Dollar, which is an economic bargaining game. The players produced fair bids with high bid totals, which is a desirable outcome, but no attempt was made to understand why the players performed so well. In this paper, we interpret the behavior of those neural networks using SHapley Additive exPlanations (or SHAP). Surprisingly, the neural network players exhibited both altruistic and exploitative behavior. Both a global and a local interpretation analysis is presented. The experiments conducted in this work demonstrate a simple method for understanding players’ strategies in multi-player games.

Keywords: Deep Learning
Abstract: Neural Architecture Search (NAS) aims to automate the creation of Artificial Neural Networks, including Convolutional Neural Networks (CNN), lessening the reliance on labour-intensive manual design by human experts. A CNN architecture can be decomposed into a micro- and macro-architecture, each influenced by distinct design and optimisation strategies to contribute to the overall construction and performance of the CNN. Cellular Encoding (CE), an evolutionary computation technique, has been successfully used to represent diverse network topologies of varying complexities. Recently, CE has been applied to evolve CNN architectures, showing promising results. However, current CE-based NAS approaches focus on evolving either the micro- or macro-architectures without considering the evolution of both in the same algorithm. Evolving the micro- and macro-architecture together can increase the performance of evolved CNN architectures. This research introduces a novel two-stage hybrid approach, combining Genetic Algorithms (GA) and CE to evolve both the micro- and macro-architectures to synthesise CNNs for classification tasks. Candidate macro-architectures are evolved using a CE approach, while a GA approach is used to explore the micro-architecture search space. The proposed algorithm is evaluated across four commonly used datasets and compared against six NAS peer competitors and five state-of-the-art manually designed CNN architectures. The results validate the approach's high competitiveness, outperforming several peer competitors on image and text classification tasks.

Keywords: Evolving Learning, Advanced Optimization, Randomized Algorithms
Abstract: Remarkable successes in deep learning have spurred significant growth in the field of neural architecture search (NAS), which is rapidly advancing as a promising technique for automating the design of network architecture. From an optimization standpoint, a NAS task for a given search space can be viewed as a multi-objective optimization problem (MOP) when considering multiple design criteria simultaneously (e.g., prediction accuracy, architecture complexity, hardware efficiency). However, whether a NAS problem is a multimodal multi-objective optimization problem or not (i.e., whether a single non-dominated solution in the objective space has multiple different neural network architectures or not) has not been examined in the literature. This presents an intriguing research question that merits further investigation. To fill this gap, we examine the multimodal nature of seven multi-objective NAS problems. By doing so, this work aims to help MOP researchers to better understand the characteristics of the multi-objective NAS problems.

Keywords: Evolving Learning, Evolvable Systems, Deep Learning
Abstract: In recent years, there has been a rise in popularity of using evolutionary algorithms (EA's) in conjunction with artificial neural networks (ANN's). This approach is commonly known as NeuroEvolution (NE). NeuroEvolutionary approaches typically optimize just the weights of an ANN or optimize the architecture, learning rates, thresholds, and weights together. Algorithms capable of the latter are known as Topological and Weight Evolving ANN (TWEANN). One such TWEANN is Gene Expression Programming for Neural Networks (GEPNN). This paper presents an empirical investigation of the network topologies that arise when GEPNN is used and whether evolved architectures have any relation to state of the art architectures. Results show that GEPNN naturally discovers powerful structural motifs such as shortcut connections and also creates sparse networks. Both these schemas have been shown to be advantageous in deep learning techniques. As an additional contribution from this work, we provide an open source library for developing GEPNN solutions in Python.

Keywords: Evolving Learning, Automated Algorithm, Deep Learning
Abstract: Recent studies have shown that the performance of evolutionary neural architecture search (i.e., neuroevolution) algorithms can be significantly improved by the use of island based strategies which periodically experience extinction and repopulation events. Further, it has been shown that the simplex hyperparameter optimization (SHO) method can also improve neuroevolution (NE) performance by optimizing neural network training hyperparameters while the NE algorithm also trains and designs neural networks. This work provides an extensive examination of combining island repopulation events with five different island-based variations of SHO. These methods are evaluated for the evolution of recurrent neural networks for the challenging problem of multivariate time series forecasting on two real world datasets. We show with statistical significance that adding repopulation to the SHO variants in almost every case improves performance, and for those that does there is no statistical difference. In addition, we find that one variant in particular, multi-island, random island best genome (MIRIB) performs the best across all experiment types.