In recent years, artificial intelligence systems for arrhythmia detection in electrocardiogram (ECG) signals have achieved remarkable performance, supporting the development of automated tools for clinical time-series analysis. However, traditional machine learning approaches usually rely on handcrafted features, reducing inherently spatio-temporal signals to static representations. Although deep learning architectures can overcome this limitation by learning representations directly from data, they often entail high computational costs, large training requirements, and complex optimization procedures. These limitations motivate the exploration of alternative computational paradigms capable of preserving temporal information while reducing training complexity. Liquid State Machines (LSMs), a paradigm within Reservoir Computing (RC), offer such an alternative by projecting temporal inputs into dynamic, high-dimensional representations through a recurrent, task-independent liquid or reservoir, while learning is confined to a simple task-dependent readout. This study investigates the design choices involved in an LSM-based ECG heartbeat classification system, emphasizing its pipeline components, and practical challenges associated with applying RC-based models to biomedical time-series data. A comprehensive experimental analysis of the LSM-based pipeline is conducted, including signal preprocessing, encoding strategies, and readout configurations. Beyond performance evaluation, this work examines how these pipeline components influence the behavior, consistency, and practical trade-offs of LSM-based ECG classification. The results provide insights into the opportunities and limitations of LSM architectures for clinical time-series processing, contributing to the understanding of RC methods for physiological signal analysis.

In this paper, we propose an optimization approach using Particle Swarm Optimization (PSO) to enhance reservoir separability in Liquid State Machines (LSMs) for spatio-temporal classification in neuromorphic systems. By leveraging PSO, our method fine-tunes reservoir parameters, neuron dynamics, and connectivity patterns, maximizing separability while aligning with the resource constraints typical of neuromorphic hardware. This approach was validated in both software (NEST) and on neuromorphic hardware (SpiNNaker), demonstrating notable results in terms of accuracy and low energy consumption when using SpiNNaker. Specifically, our approach addresses two problems: Frequency Recognition (FR) with five classes and Pattern Recognition (PR) with four, eight, and twelve classes. For instance, in the Mono-objective approach running in NEST, accuracies ranged from 81.09% to 95.52% across the benchmarks under study. The Multi-objective approach outperformed the Mono-objective approach, delivering accuracies ranging from 90.23% to 98.77%, demonstrating its superior scalability for LSM implementations. On the SpiNNaker platform, the mono-objective approach achieved accuracies ranging from 86.20% to 97.70% across the same benchmarks, with the Multi-objective approach further improving accuracies, ranging from 94.42% to 99.52%. These results show that, in addition to slight accuracy improvements, hardware-based implementations offer superior energy efficiency with a lower execution time. For example, SpiNNaker operates at around 1–5 watts per chip, while traditional systems can require 50–100 watts for similar tasks, highlighting the significant energy savings of neuromorphic hardware. These results underscore the scalability and effectiveness of PSO-optimized LSMs on resource-limited neuromorphic platforms, showcasing both improved classification performance and the advantages of energy-efficient processing.

Ultrafast pulsed laser technology presents unique challenges and opportunities in material processing and characterization for precision photonics. Herein, an experiment is conducted involving the use of an ultrafast pulsed laser to irradiate a molybdenum film, inducing oxide formation. A total of 54 experiments are performed, varying the laser irradiation time and per‐pulse laser fluence, resulting in a database with diverse oxide formations on the material. This dataset is further expanded numerically through interpolation to 187 samples. Subsequently, eight different deep neural network models, each with varying hidden layers and numbers of neurons, are employed to characterize the laser behavior with different parameters. These models are then validated numerically using three different learning rates, and the results are statistically evaluated using three metrics: mean squared error, mean absolute error, and R2 score.

In this work, we propose the integration of a mechanism to enable smooth transitions between different locomotion patterns in a hexapod robot. Specifically, we utilize a spiking neural network (SNN) functioning as a Central Pattern Generator (CPG) to generate three distinct locomotion patterns, or gaits: walk, jog, and run. This network produces coordinated spike trains, mimicking those generated in the brain, which are translated into synchronized robot movements via PWM signals. Subsequently, these spike trains are compared using a similarity metric known as SPIKE-synchronization to identify the optimal point for transitioning from one gait to another. This approach aims to achieve three main objectives: first, to maintain the robot’s balance during transitions; second, to ensure that gait transitions are almost imperceptible; and third, to improve energy efficiency by reducing abrupt changes in the robot’s actuators (servomotors). To validate our proposal, we incorporated FSR sensors on the robot’s legs to detect the rigidity of the terrain it navigates. Based on the terrain’s rigidity, the robot dynamically transitions between gaits. The system was tested in real time on a physical hexapod robot across four different types of terrain. Although the method was validated exclusively on a hexapod robot, it can be extended to any legged robot.

In this paper, we present an evolutionary design methodology of Central Pattern Generator (CPG)-based locomotion systems for hexapod and quadrupedal robots. The CPGs are built as Spiking Neural Networks, whose synaptic connections and weights are directly configured by an evolutionary algorithm in order that CPGs generate rhythmic and periodical signals to carry out robotic locomotion. The CPGs are fully designed and implemented using Nengo simulator. There were obtained CPGs for different locomotion patterns for both, hexapod and quadruped robots. The obtained CPGs achieve behaviours reported in the state of the art with smaller architectures.

In real-world classification tasks, it is common to find class imbalance issues in the training datasets, i.e. an unequal number of examples among the different classes. The class imbalance problem biases the performance of predictive models by overlooking minority classes; this is because predictive models employ learning rules with accuracy-based cost functions, thus favoring majority classes. In this work, the class imbalance issue is tackled through generative models, using the BreakHis dataset, a histopathologic image set intended for breast cancer classification, as a case study. The BreasHis' minority class is balanced by adding synthetic images obtained by means of different generative methods, including variational autoencoders and two different generative adversarial networks. The quality of the image sets created by the different generative models, and their effects in balancing the BreakHis dataset, are evaluated through several quantitative metrics computed from classification tasks. Statistical analysis is performed and the results indicate that the DCGAN network is superior to the other evaluated models.

Grammar Evolution (GE) can be considered a form of Genetic Programming (GP) that has become very popular in the field of Automatic Programming (AP) over the last few years. There has been a lot of research on different aspects of GE, including its parts; the Search Engine, Mapping Process, and Grammar. However, it has been shown that it is possible to select the codons randomly to improve the GE, using a random permutation. This paper introduces a new approach to intensify a solution using permutation heuristics to guide the codon selection order. A non -parametric test was applied to discern between the results obtained by the proposal and those obtained by the canonical GE version and the GE with random permutations.

The Liquid State Machine (LSM) framework addresses supervised learning tasks involving spatio-temporal data streams. It relies on a randomly created, untrained Spiking Recurrent Neural Network (SRNN), called the “liquid,” to map inputs into task-independent representations. A simple readout layer then uses these representations to solve specific tasks. LSM’s computational power arises from two properties: the Separation Property (related to the liquid) and the Approximation Property (related to the readout). This research aims to enhance the liquid’s separation property to improve classification performance and enable multitask learning through swarm intelligence. The study develops a two-phase approach: first, using Particle Swarm Optimization (PSO) to optimize the liquid for distinguishing data streams of different classes in single tasks; and second, extending this optimization to multitask learning with Original Multi-Objective PSO (OMOPSO). Results from experiments on four artificial problems (one of frequency recognition and three of pattern recognition) demonstrate that optimized liquids improve separability and maintain regularized firing behaviors, even with a simple softmax readout layer. On average, the experiments show that our approach outperforms baseline methods across all four artificial datasets when using PSO and achieves superior results on three pattern recognition datasets when employing OMOPSO.

Online hyper-heuristics are algorithms capable of solving complex real-world problems. This approach dynamically selects, based on the quality of a given solution state, the most promising operator from a pool to continue a search process. In multi-objective optimization problems, more than one objective function has to be optimized simultaneously. Many multi-objective optimization problems commonly represent complex real-world problems. Despite the success of online hyper-heuristics applied to single-objective problems, there is a lack of work on Hyper-heuristics applied to multi-objective optimization. In this paper, we propose an approach to deal with multi-objective optimization with an online hyper-heuristic using as a quality metric, the Hypervolume (Hv) indicator, in a MOEA/D as a high-Level algorithm. The Hypervolume metric is a set measure used in multi-objective optimization to evaluate the performance of a search process transforming multiple objective problems into single-objective ones. This allows us to apply a single objective online hyper-heuristic to multi-objective problems. For experimentation, we use the well-known ZDT and DTLZ benchmarks.

Real-world optimization problems can use a high number of computational resources to be solved because of the complexity of the fitness function or objective function. This situation negatively impacts the performance of any Heuristic or deterministic solver since most of the computational time is invested into processing the fitness function. Surrogate models are used to reduce the computational time of processing complex functions by replacing extensive numerical simulations with approximate functions that are faster to evaluate. Surrogate models are constructed by evaluating the original model at a set of points (training), and then using the corresponding evaluations to construct an approximate model based on given mathematical functions. The main idea is to have a faster function that can guide an optimization solver with a minimum error or difference against the original fitness function. In this work, several surrogated models are constructed and compared in the context of a heuristic optimization solver to analyze their performance as fitness functions.

Currently, COVID-19, Influenza, and Dengue are diseases of international concern that have endured over time and affected various regions of the world at educational, economic, social, and health levels. Research efforts in computer science have focused on curve-fitting approaches to predict the behavior of an epidemic. However, the mathematical characterization of epidemic spread is a subject little explored by these techniques. In this work, we propose to use an artificial intelligence system that implements computational optimization and polynomial adjustment techniques called META-COVID19 to mathematically characterize the spread of epidemics in Mexico. The results of our research show the explicit mathematical model that characterizes the time series, which are likely to be analyzed by formal mathematical methods to obtain valuable information on the epidemiological behavior of different diseases of interest.

Brain-computer interfaces (BCI) are a mechanism to record the electrical signals of the brain and translate them into commands to operate an output device like a robotic system. This article presents the development of a real-time locomotion system of a hexapod robot with bio-inspired movement dynamics inspired in the stick insect and tele-operated by cognitive activities of motor imagination. Brain signals are acquired using only four electrodes from a BCI device and sent to computer equipment for processing and classification by the iQSA method based on quaternion algebra. A structure consisting of three main stages are proposed: (1) signal acquisition, (2) data analysis and processing by the iQSA method, and (3) bio-inspired locomotion system using a Spiking Neural Network (SNN) with twelve neurons. An off-line training stage was carried out with data from 120 users to create the necessary decision rules for the iQSA method, obtaining an average performance of 97.72%. Finally, the experiment was implemented in real-time to evaluate the performance of the entire system. The recognition rate to achieve the corresponding gait pattern is greater than 90% for BCI, and the time delay is approximately from 1 to 1.5 seconds. The results show that all the subjects could generate their desired mental activities, and the robotic system could replicate the gait pattern in line with a slight delay.

A micro-evolutionary algorithm based on differential evolution (MiDE) is used to carry out a fast and effective exploration of the conformational space of molecular clusters (with dimensions about 8 Å), which can be considered nanoparticle or nanoparticle scaffolds, finding and optimizing the minimum energy conformers. The search of the best geometry takes advantage of MiDE to perform good exploration over search space starting with a few candidate solutions. The presented software in this contribution, Micro-differential evolution cluster-optimizer (MiDECO), automatizes such exploration process. The Gaussian 09 package is used to perform the geometric optimizations of the clusters to compare the different conformations through the computed energies, which allowed the most stable molecules to be discerned. The software presented works for clusters of tested sets, reproducing the results reported in the state of the art and obtaining a significant reduction of the computational cost due to the micro-population used with the MiDE without using high-performance computing.