Search

Ewing's sarcoma (ES), characterized by a high density of small round blue cells without structural organization, presents a significant health concern, particularly among adolescents aged 10 to 19. Artificial intelligence-based systems for automated analysis of histopathological images are promising to contribute to an accurate diagnosis of ES. In this context, this study explores the feature extraction ability of different pre-training strategies for distinguishing ES from other soft tissue or bone sarcomas with similar morphology in digitized tissue microarrays for the first time, as far as we know. Vision-language supervision (VLS) is compared to fully-supervised ImageNet pre-training within a multiple instance learning paradigm. Our findings indicate a substantial improvement in diagnostic accuracy with the adaption of VLS using an in-domain dataset. Notably, these models not only enhance the accuracy of predicted classes but also drastically reduce the number of trainable parameters and computational costs., Comment: 11 pages, 5 figures, 2 tables. Oral presentation at KES-InMed 2024 held in Madeira, Portugal

We show that for every g greater or equal than 5, the locus of Prym varieties in the moduli space of principally polarized abelian varieties of dimension g-1 that possess a pseudoreflection of geometric origin is the union of three different non-empty explicit irreducible families. This is in stark contrast to the loci of Jacobian varieties that possess a pseudoreflection of geometric origin, which is empty for any genus greater than 3. In g=6, a distinguished example of Prym varieties with a pseudoreflection is given by intermediate Jacobians of cubic threefolds that possess an Eckardt point., Comment: 17 pages, comments are welcome

The integration of conversational agents into our daily lives has become increasingly common, yet many of these agents cannot engage in deep interactions with humans. Despite this, there is a noticeable shortage of datasets that capture multimodal information from human-robot interaction dialogues. To address this gap, we have developed a Personal Emotional Robotic Conversational sYstem (PERCY) and recorded a novel multimodal dataset that encompasses rich embodied interaction data. The process involved asking participants to complete a questionnaire and gathering their profiles on ten topics, such as hobbies and favourite music. Subsequently, we initiated conversations between the robot and the participants, leveraging GPT-4 to generate contextually appropriate responses based on the participant's profile and emotional state, as determined by facial expression recognition and sentiment analysis. Automatic and user evaluations were conducted to assess the overall quality of the collected data. The results of both evaluations indicated a high level of naturalness, engagement, fluency, consistency, and relevance in the conversation, as well as the robot's ability to provide empathetic responses. It is worth noting that the dataset is derived from genuine interactions with the robot, involving participants who provided personal information and conveyed actual emotions., Comment: 9 pages, 5 Figures, Rejected from International Conference of Human Robot Interaction 2025, Melbourne, Australia

We report on an educational pilot program for low-cost physics experimentation run in Ecuador, South Africa, and the United States. The program was developed after having needs-based discussions with African educators, researchers, and leaders. It was determined that the need and desire for low-cost, skills-building, and active-learning tools is very high. From this, we developed a 3D-printable, Raspberry Pi-based multispectral camera (15 to 25 spectral channels in the visible and near-IR) for as little as $100. The program allows students to learn 3D modeling, 3D printing, feedback, control, image analysis, Python programming, systems integration and artificial intelligence as well as spectroscopy. After completing their cameras, the students in the program studied plant health, plant stress, post-harvest fruit ripeness, and polarization and spectral analysis of nanostructured insect wings, the latter of which won the ``best-applied research" award at a conference poster session and will be highlighted in this paper. Importantly, these cameras can be an integral part of any developing country's agricultural, recycling, medical, and pharmaceutical infrastructure. Thus, we believe this experiment can play an important role at the intersection of student training and developing countries' capacity building.

Domain shift in the field of histopathological imaging is a common phenomenon due to the intra- and inter-hospital variability of staining and digitization protocols. The implementation of robust models, capable of creating generalized domains, represents a need to be solved. In this work, a new domain adaptation method to deal with the variability between histopathological images from multiple centers is presented. In particular, our method adds a training constraint to the supervised contrastive learning approach to achieve domain adaptation and improve inter-class separability. Experiments performed on domain adaptation and classification of whole-slide images of six skin cancer subtypes from two centers demonstrate the method's usefulness. The results reflect superior performance compared to not using domain adaptation after feature extraction or staining normalization., Comment: Accepted in CASEIB 2024

We study an explicit $(2g-1)$-dimensional family of Jacobian varieties of dimension $\frac{d-1}2(g-1)$, arising from quotient curves of unramified cyclic coverings of prime degree $d$ of hyperelliptic curves of genus $g\ge 2$. By using a deformation argument, we prove that the generic element of the family is simple. Furthermore, we completely describe their endomorphism algebra, and we show that they admit a rank $\frac{d-1}2-1$ group of non-polarized automorphisms. As an application of these results, we prove the generic injectivity of the Prym map for \'etale cyclic coverings of hyperelliptic curves of odd prime degree under some slight numerical restrictions. This result generalizes in several directions previous results on genus 2.

Despite being the subject of study for several years, excessive vibration persists in the machining of metal parts. In this context, the Stability Lobe Diagram (SLD) is presented as a viable tool to mitigate this problem as a function of axial depth of cut and spindle speed. However, its accurate construction is subject to the consideration of multiple parameters and models, whose application may be affected by certain inherent uncertainties. In turn, this impacts its accuracy, especially in the stability and instability regions. The present study aims to characterize these uncertainties, analyze their influence on the SLD, and propose strategies for their reduction. Ultimately, the goal is to facilitate the user's decision-making when choosing the trajectory generation parameters.

Determining the underlying equations of motion and parameter values for vibrating structures is of great concern in science and engineering. This work introduces a new data-driven approach called the energy-based dual-phase dynamics identification (EDDI) method for identifying the nonlinear dynamics of single-degree-of-freedom oscillators. The EDDI method leverages the energies of the system to identify the governing dynamics through the forces acting on the oscillator. The approach consists of two phases: a model-dissipative and model-stiffness identification. In the first phase, the fact that kinetic and mechanical energies are equivalent when the displacement is zero is leveraged to compute the energy dissipated and a corresponding model for the nonlinear damping of the system. In the second phase, the energy dissipated is used to compute the mechanical energy (ME), which is then used to obtain a reformulated Lagrangian. The conservative forces acting on the oscillator are then computed by taking the derivative the Lagrangian, then a model for the nonlinear stiffness is identified by solving a system of linear equations. The resulting governing equations are identified by including both the nonlinear damping and stiffness terms. A key novelty of the EDDI method is that the only thing required to perform the identification is free-response measurements and the mass of the oscillator. No prior understanding of the dynamics of the system is necessary to identify the underlying dynamics, such that the EDDI method is a truly data-driven method. The method is demonstrated using simulated and measured responses of nonlinear single-degree-of-freedom systems with a variety of nonlinear mechanisms.

Identifying nonlinear dynamics and characterizing noise from data is critical across science and engineering for understanding and modeling the behavior of the systems accurately. The modified sparse identification of nonlinear dynamics (mSINDy) has emerged as an effective framework for identifying systems embedded in heavy noise; however, further improvements can expand its capabilities and robustness. By integrating the weak SINDy (WSINDy) into mSINDy, we introduce the weak mSINDy (WmSINDy) to improve the system identification and noise modeling by harnessing the strengths of both approaches. The proposed algorithm simultaneously identifies parsimonious nonlinear dynamics and extracts noise probability distributions using automatic differentiation. We evaluate WmSINDy using several nonlinear systems and it demonstrates improved accuracy and noise characterization over baselines for systems embedded in relatively strong noise., Comment: Submitted to Journal of Computational Physics

Since the emergence of the ImageNet dataset, the pretraining and fine-tuning approach has become widely adopted in computer vision due to the ability of ImageNet-pretrained models to learn a wide variety of visual features. However, a significant challenge arises when adapting these models to domain-specific fields, such as digital pathology, due to substantial gaps between domains. To address this limitation, foundation models (FM) have been trained on large-scale in-domain datasets to learn the intricate features of histopathology images. In cancer diagnosis, whole-slide image (WSI) prediction is essential for patient prognosis, and multiple instance learning (MIL) has been implemented to handle the giga-pixel size of WSI. As MIL frameworks rely on patch-level feature aggregation, this work aims to compare the performance of various feature extractors developed under different pretraining strategies for cancer subtyping on WSI under a MIL framework. Results demonstrate the ability of foundation models to surpass ImageNet-pretrained models for the prediction of six skin cancer subtypes, Comment: Manuscript accepted for oral presentation at Decision Science Allieance -INternational Summer Conference (DSA-ISC) 2024 held on Valencia, Spain

Vision-language supervision has made remarkable strides in learning visual representations from textual guidance. In digital pathology, vision-language models (VLM), pre-trained on curated datasets of histological image-captions, have been adapted to downstream tasks, such as region of interest classification. Zero-shot transfer for slide-level prediction has been formulated by MI-Zero, but it exhibits high variability depending on the textual prompts. Inspired by prototypical learning, we propose MI-VisionShot, a training-free adaptation method on top of VLMs to predict slide-level labels in few-shot learning scenarios. Our framework takes advantage of the excellent representation learning of VLM to create prototype-based classifiers under a multiple-instance setting by retrieving the most discriminative patches within each slide. Experimentation through different settings shows the ability of MI-VisionShot to surpass zero-shot transfer with lower variability, even in low-shot scenarios. Code coming soon at thttps://github.com/cvblab/MIVisionShot., Comment: Manuscript accepted for oral presentation at KES-InnovationInMedicine 2024 held on Madeira, Portugal

In this article, we study the complexity of the upgrading version of the maximal covering location problem with edge length modifications on networks. This problem is NP-hard on general networks. However, in some particular cases, we prove that this problem is solvable in polynomial time. The cases of star and path networks combined with different assumptions for the model parameters are analysed. In particular, we obtain that the problem on star networks is solvable in O(nlogn) time for uniform weights and NP-hard for non-uniform weights. On paths, the single facility problem is solvable in O(n^3) time, while the p-facility problem is NP-hard even with uniform costs and upper bounds (maximal upgrading per edge), as well as, integer parameter values. Furthermore, a pseudo-polynomial algorithm is developed for the single facility problem on trees with integer parameters.

We study the upgrading version of the maximal covering location problem with edge length modifications on networks. This problem aims at locating p facilities on the vertices (of the network) so as to maximise coverage, considering that the length of the edges can be reduced at a cost, subject to a given budget. Hence, we have to decide on: the optimal location of p facilities and the optimal edge length reductions. This problem is NP-hard on general graphs. To solve it, we propose three different mixed-integer formulations and a preprocessing phase for fixing variables and removing some of the constraints. Moreover, we strengthen the proposed formulations including valid inequalities. Finally, we compare the three formulations and their corresponding improvements by testing their performance over different datasets.

This paper addresses a version of the single-facility Maximal Covering Location Problem on a network where the demand is: (i) distributed along the edges and (ii) uncertain with only a known interval estimation. To deal with this problem, we propose a minmax regret model where the service facility can be located anywhere along the network. This problem is called Minmax Regret Maximal Covering Location Problem with demand distributed along the edges (MMR-EMCLP). Furthermore, we present two polynomial algorithms for finding the location that minimises the maximal regret assuming that the demand realisation is an unknown constant or linear function on each edge. We also include two illustrative examples as well as a computational study for the unknown constant demand case to illustrate the potential and limits of the proposed methodology.

We explore the Schmidt-Kennicutt (SK) relations and the star formation efficiency per free-fall time ($\eff$), mirroring observational studies, in numerical simulations of filamentary molecular clouds undergoing gravitational contraction. We find that {\it a)} collapsing clouds accurately replicate the observed SK relations for galactic clouds and {\it b)} the so-called efficiency per free-fall time ($\eff$) is small and constant in space and in time, with values similar to those found in local clouds. This constancy is a consequence of the similar radial scaling of the free-fall time and the internal mass in density structures with spherically-averaged density profiles near $r^{-2}$. We additionally show that {\it c)} the star formation rate (SFR) increases rapidly in time; {\it d)} the low values of $\eff$ are due to the different time periods over which $\tauff$ and $\tausf$ are evaluated, together with the fast increasing SFR, and {\it e)} the fact that star clusters are significantly denser than the gas clumps from which they form is a natural consequence of the fast increasing SFR, the continuous replenishment of the star-forming gas by the accretion flow, and the near $r^{-2}$ density profile generated by the collapse Finally, we argue that the interpretation of $\eff$ as an efficiency is problematic because its maximum value is not bounded by unity, and because the total gas mass in the clouds is not fixed, but rather depends on the environment where clouds are embedded. In summary, our results show that the SK relation, the typical observed values of $\eff$, and the mass density of clusters arise as a natural consequence of gravitational contraction., Comment: Submitted to MNRAS

The paper presents an extension of the metaphysics of self-subsisting structures set out in a companion paper to the realm of non-relativistic quantum physics. The discussion is centered around a Pure Shape Dynamics model representing a relational implementation of a de Broglie-Bohm $N$-body system. An interpretation of this model in terms of self-subsisting structures is proposed and assessed against the background of the debate on the metaphysics of quantum physics, with a particular emphasis on the nature of the wave function. The analysis shows that elaborating an appropriate Leibnizian/Machian metaphysics of the quantum world requires a substantial revision of the notion of world-building relation., Comment: 32 pages, forthcoming in Foundations of Physics

We introduce a class of examples which provide an affine generalization of the nonholonomic problem of a convex body rolling without slipping on the plane. We investigate dynamical aspects of the system such as existence of first integrals, smooth invariant measure and integrability, giving special attention to the cases in which the convex body is a dynamically balanced sphere or a body of revolution., Comment: 29 pages, 8 figures

The topics of interest are location analysis and related problems. This includes location models, networks, transportation, logistics, exact and heuristic solution methods, and computational geometry, among many others., Comment: The proceedings book of the previous editions can be found at arXiv:2002.08287 arXiv:2002.08293 arXiv:2002.08300 arXiv:2002.01702 arXiv:2202.13878 . arXiv:2309.08337 . The XIII International Workshop on Locational Analysis and Related Problems will take place during September 4--6, 2024 in Granada (Spain). It is organized by the Spanish Location Network, REDLOCA, and the Location Group, GELOCA, from the Spanish Society of Statistics and Operations Research (SEIO). The Spanish Location Network is a group of 100+ researchers from several Spanish universities organized into 7 thematic groups. The Network has been funded by the Spanish Government since 2003

Photons emitted from high-energy electron beam interactions with high-field systems, such as the upcoming FACET-II experiments at SLAC National Accelerator Laboratory, may provide deep insight into the electron beam's underlying dynamics at the interaction point. With high-energy photons being utilized to generate electron-positron pairs in a novel spectrometer, there remains a key problem of interpreting the spectrometer's raw data to determine the energy distribution of the incoming photons. This paper uses data from simulations of the primary radiation emitted from electron interactions with a high-field, short-pulse laser to determine optimally reliable methods of reconstructing the measured photon energy distributions. For these measurements, recovering the emitted 10 MeV to 10 GeV photon energy spectra from the pair spectrometer currently being commissioned requires testing multiple methods to finalize a pipeline from the spectrometer data to incident photon and, by extension, electron beam information. In this study, we compare the performance QR decomposition, a matrix deconstruction technique and neural network with and without maximum likelihood estimation (MLE). Although QR decomposition proved to be the most effective theoretically, combining machine learning and MLE proved to be superior in the presence of noise, indicating its promise for analysis pipelines involving high-energy photons., Comment: 11 Pages, 12 Figures. arXiv admin note: substantial text overlap with arXiv:2209.12119

We report relaxation oscillations in a one-dimensional array of Josephson junctions. The oscillations are circuit-dual to those ordinarily observed in single junctions. The dual circuit quantitatively accounts for temporal dynamics of the array, including the dependence on biasing conditions. Injection locking the oscillations results in well-developed current plateaux. A thermal model explains the relaxation step of the oscillations., Comment: 7 pages, 8 figures

Intent detection is a text classification task whose aim is to recognize and label the semantics behind a users query. It plays a critical role in various business applications. The output of the intent detection module strongly conditions the behavior of the whole system. This sequence analysis task is mainly tackled using deep learning techniques. Despite the widespread use of these techniques, the internal mechanisms used by networks to solve the problem are poorly understood. Recent lines of work have analyzed the computational mechanisms learned by RNNs from a dynamical systems perspective. In this work, we investigate how different RNN architectures solve the SNIPS intent detection problem. Sentences injected into trained networks can be interpreted as trajectories traversing a hidden state space. This space is constrained to a low-dimensional manifold whose dimensionality is related to the embedding and hidden layer sizes. To generate predictions, RNN steers the trajectories towards concrete regions, spatially aligned with the output layer matrix rows directions. Underlying the system dynamics, an unexpected fixed point topology has been identified with a limited number of attractors. Our results provide new insights into the inner workings of networks that solve the intent detection task., Comment: Camera-Ready version. Accepted paper at 27th European Conference on Artificial Intelligence (ECAI-2024)

The paper discusses the possible implications of the relational framework of Pure Shape Dynamics for the metaphysics of time. The starting point of the analysis is an interpretation of shapes in ontic structural realist terms, which gives rise to the notion of self-subsisting structure. The relational version of a Newtonian-particle toy model is introduced and discussed as a concrete example., Comment: 25 pages, 3 figures. Forthcoming in the special issue "On time in the foundations of physics" of the Journal for General Philosophy of Science

The Prym map $\mathcal{P}_6$ in genus 6 is dominant and generically finite of degree 27. When restricted to the divisor of curves with an odd semicanonical pencil $\mathcal{T}_6^o$, it is still generically finite, but of degree strictly smaller. In this paper, we prove that $\mathcal{P}_6$ restricted to $\mathcal{T}_6^o$ is birational and that the monodromy group over the image of $\mathcal{T}_6^o$ is the Weyl group $WD_5$. Thus, there are two other irreducible divisors in the moduli space of Prym curves $\mathcal{R}_6$ and the degree of $\mathcal{P}_6$ restricted to them is 10 and 16. Moreover, we study the geometry of the divisor where $\mathcal{P}_6$ has degree 10., Comment: 16 pages, comments are welcome

In the underdense (blowout) regime of plasma wakefield acceleration (PWFA), the particle beam is denser than the plasma. Under these conditions, the plasma electrons are nearly completely rarefacted from the beam channel, resulting in a nominally uniform ion column. Extensive investigations of this interaction assuming axisymmetry have been undertaken. However, the plasma blowout produced by a transversely asymmetric driver possesses quite different characteristics. They create an asymmetric plasma rarefaction region (bubble) which leads to asymmetric focusing in the two transverse planes. This is also accompanied by an undesired non-uniform accelerating gradient. The asymmetric blowout cross-section is found through simulation to be elliptical, and treating it as such permits a simple extension of the symmetric theory. In particular, focusing fields linear in both transverse directions exist in the bubble. The form of the wake potential and the concomitant matching conditions in this elliptical cavity are discussed in this paper. We also discuss bubble boundary estimation in the long driver limit and applications of the asymmetric features of the wakefield., Comment: 8 pages, 4 figures

Insect pest control poses a global challenge, affecting public health, food safety, and the environment. Diseases transmitted by mosquitoes are expanding beyond tropical regions due to climate change. Agricultural pests further exacerbate economic losses by damaging crops. The Sterile Insect Technique (SIT) emerges as an eco-friendly alternative to chemical pesticides, involving the sterilization and release of male insects to curb population growth. This work focuses on the automation of the analysis of field ovitraps used to follow-up a SIT program for the Aedes albopictus mosquito in the Valencian Community, Spain, funded by the Conselleria de Agricultura, Agua, Ganaderia y Pesca. Previous research has leveraged deep learning algorithms to automate egg counting in ovitraps, yet faced challenges such as manual handling and limited analysis capacity. Innovations in our study include classifying eggs as hatched or unhatched and reconstructing ovitraps from partial images, mitigating issues of duplicity and cut eggs. Also, our device can analyze multiple ovitraps simultaneously without the need of manual replacement. This approach significantly enhances the accuracy of egg counting and classification, providing a valuable tool for large-scale field studies. This document describes part of the work of the project Application of Industry 4.0 techniques to the production of tiger mosquitoes for the Sterile Insect Technique (MoTIA2,IMDEEA/2022/70), financed by the Valencian Institute for Business Competitiveness (IVACE) and the FEDER funds. The participation of J.Naranjo-Alcazar, J.Grau-Haro and P.Zuccarello has been possible thanks to funding from IVACE and FEDER funds. The participation of D.Almenar has been financed by the Conselleria de Agricultura, Agua, Ganaderia y Pesca of the Generalitat Valenciana and the Subdireccion de Innovacion y Desarrollo de Servicios (TRAGSA group).

Machine Listening focuses on developing technologies to extract relevant information from audio signals. A critical aspect of these projects is the acquisition and labeling of contextualized data, which is inherently complex and requires specific resources and strategies. Despite the availability of some audio datasets, many are unsuitable for commercial applications. The paper emphasizes the importance of Active Learning (AL) using expert labelers over crowdsourcing, which often lacks detailed insights into dataset structures. AL is an iterative process combining human labelers and AI models to optimize the labeling budget by intelligently selecting samples for human review. This approach addresses the challenge of handling large, constantly growing datasets that exceed available computational resources and memory. The paper presents a comprehensive data-centric framework for Machine Listening projects, detailing the configuration of recording nodes, database structure, and labeling budget optimization in resource-constrained scenarios. Applied to an industrial port in Valencia, Spain, the framework successfully labeled 6540 ten-second audio samples over five months with a small team, demonstrating its effectiveness and adaptability to various resource availability situations. Acknowledgments: The participation of Javier Naranjo-Alcazar, Jordi Grau-Haro and Pedro Zuccarello in this research was funded by the Valencian Institute for Business Competitiveness (IVACE) and the FEDER funds by means of project Soroll-IA2 (IMDEEA/2023/91). The research carried out for this publication has been partially funded by the project STARRING-NEURO (PID2022-137048OA-C44) funded by the Ministry of Science, Innovation and Universities of Spain and the European Union., Comment: Paper accepted at 8th Future of Information and Communication Conference 2025, 28-29 April, Berlin

We present a pioneering investigation into the application of deep learning techniques to analyze histopathological images for addressing the substantial challenge of automated prognostic prediction. Prognostic prediction poses a unique challenge as the ground truth labels are inherently weak, and the model must anticipate future events that are not directly observable in the image. To address this challenge, we propose a novel three-part framework comprising of a convolutional network based tissue segmentation algorithm for region of interest delineation, a contrastive learning module for feature extraction, and a nested multiple instance learning classification module. Our study explores the significance of various regions of interest within the histopathological slides and exploits diverse learning scenarios. The pipeline is initially validated on artificially generated data and a simpler diagnostic task. Transitioning to prognostic prediction, tasks become more challenging. Employing bladder cancer as use case, our best models yield an AUC of 0.721 and 0.678 for recurrence and treatment outcome prediction respectively., Comment: https://github.com/Biomedical-Data-Analysis-Laboratory/HistoPrognostics

In Evolutionary Game Theory (EGT), a population reaches a Nash equilibrium when none of the agents can improve its objective by solely changing its strategy on its own. Roughly speaking, this equilibrium is a protection against betrayal. Generalized Nash Equilibrium (GNE) is a more complex version of this idea with important implications in real-life problems in economics, wireless communication, the electricity market, or engineering among other areas. In this paper, we propose a first approach to GNE with Membrane Computing techniques and show how GNE problems can be modeled with P systems, bridging both areas and opening a door for a flow of problems and solutions in both directions.

We present some rigorous results on the absence of a wide class of invariant measures for dynamical systems possessing attractors. We then consider a generalization of the classical nonholonomic Suslov problem which shows how previous investigations of existence of invariant measures for nonholonomic systems should necessarily be extended beyond the class of measures with strictly positive $C^1$ densities if one wishes to determine dynamical obstructions to the presence of attractors., Comment: 13 pages, 3 figures

This paper introduces a new hierarchical facility location model with three levels: first-level facilities which manufacture different products, second-level facilities which act as warehouses and a third-level consisting of the clients who demand the products that have been manufactured in the first level and stored in the second level. In this model, called multi-product maximal covering second-level facility location problem (SL-MCFLP), the aim is to decide the location of the second-level facilities and the products to be stored in each of them maximizing the overall clients' satisfaction with respect their coverage. To deal with this model, we introduce a Mixed Integer Linear Program (MILP) which is reinforced by some families of valid inequalities. Since some of these families have an exponential number of constraints, separation algorithms are proposed. In addition, three variants of a matheuristic procedure are developed. Computational studies are included, showing the potentials and limits of the formulation and the effectiveness of the heuristic.

In recent years, Deep Learning has gained popularity for its ability to solve complex classification tasks, increasingly delivering better results thanks to the development of more accurate models, the availability of huge volumes of data and the improved computational capabilities of modern computers. However, these improvements in performance also bring efficiency problems, related to the storage of datasets and models, and to the waste of energy and time involved in both the training and inference processes. In this context, data reduction can help reduce energy consumption when training a deep learning model. In this paper, we present up to eight different methods to reduce the size of a tabular training dataset, and we develop a Python package to apply them. We also introduce a representativeness metric based on topology to measure how similar are the reduced datasets and the full training dataset. Additionally, we develop a methodology to apply these data reduction methods to image datasets for object detection tasks. Finally, we experimentally compare how these data reduction methods affect the representativeness of the reduced dataset, the energy consumption and the predictive performance of the model.

In this paper, we present SIMAP, a novel layer integrated into deep learning models, aimed at enhancing the interpretability of the output. The SIMAP layer is an enhanced version of Simplicial-Map Neural Networks (SMNNs), an explainable neural network based on support sets and simplicial maps (functions used in topology to transform shapes while preserving their structural connectivity). The novelty of the methodology proposed in this paper is two-fold: Firstly, SIMAP layers work in combination with other deep learning architectures as an interpretable layer substituting classic dense final layers. Secondly, unlike SMNNs, the support set is based on a fixed maximal simplex, the barycentric subdivision being efficiently computed with a matrix-based multiplication algorithm.

This paper proposes a robust classification model, based on support vector machine (SVM), which simultaneously deals with outliers detection and feature selection. The classifier is built considering the ramp loss margin error and it includes a budget constraint to limit the number of selected features. The search of this classifier is modeled using a mixed-integer formulation with big M parameters. Two different approaches (exact and heuristic) are proposed to solve the model. The heuristic approach is validated by comparing the quality of the solutions provided by this approach with the exact approach. In addition, the classifiers obtained with the heuristic method are tested and compared with existing SVM-based models to demonstrate their efficiency.

This paper considers various models of support vector machines with ramp loss, these being an efficient and robust tool in supervised classification for the detection of outliers. The exact solution approaches for the resulting optimization problem are of high demand for large datasets. Hence, the goal of this paper is to develop algorithms that provide efficient methodologies to exactly solve these optimization problems. These approaches are based on three strategies for obtaining tightened values of the big M parameters included in the formulation of the problem. Two of them require solving a sequence of continuous problems, while the third uses the Lagrangian relaxation to tighten the bounds. The proposed resolution methods are valid for the l1-norm and l2-norm ramp loss formulations. They were tested and compared with existing solution methods in simulated and real-life datasets, showing the efficiency of the developed methodology.

Spontaneous datasets for Speech Emotion Recognition (SER) are scarce and frequently derived from laboratory environments or staged scenarios, such as TV shows, limiting their application in real-world contexts. We developed and publicly released the Emotional Voice Messages (EMOVOME) dataset, including 999 voice messages from real conversations of 100 Spanish speakers on a messaging app, labeled in continuous and discrete emotions by expert and non-expert annotators. We evaluated speaker-independent SER models using acoustic features as baseline and transformer-based models. We compared the results with reference datasets including acted and elicited speech, and analyzed the influence of annotators and gender fairness. The pre-trained UniSpeech-SAT-Large model achieved the highest results, 61.64% and 55.57% Unweighted Accuracy (UA) for 3-class valence and arousal prediction respectively on EMOVOME, a 10% improvement over baseline models. For the emotion categories, 42.58% UA was obtained. EMOVOME performed lower than the acted RAVDESS dataset. The elicited IEMOCAP dataset also outperformed EMOVOME in predicting emotion categories, while similar results were obtained in valence and arousal. EMOVOME outcomes varied with annotator labels, showing better results and fairness when combining expert and non-expert annotations. This study highlights the gap between controlled and real-life scenarios, supporting further advancements in recognizing genuine emotions., Comment: This article is a merged version of the description of the EMOVOME database in arXiv:2402.17496v1 and the speech emotion recognition models in arXiv:2403.02167v1. This work has been submitted to the IEEE for possible publication

Emotional Voice Messages (EMOVOME) is a spontaneous speech dataset containing 999 audio messages from real conversations on a messaging app from 100 Spanish speakers, gender balanced. Voice messages were produced in-the-wild conditions before participants were recruited, avoiding any conscious bias due to laboratory environment. Audios were labeled in valence and arousal dimensions by three non-experts and two experts, which were then combined to obtain a final label per dimension. The experts also provided an extra label corresponding to seven emotion categories. To set a baseline for future investigations using EMOVOME, we implemented emotion recognition models using both speech and audio transcriptions. For speech, we used the standard eGeMAPS feature set and support vector machines, obtaining 49.27% and 44.71% unweighted accuracy for valence and arousal respectively. For text, we fine-tuned a multilingual BERT model and achieved 61.15% and 47.43% unweighted accuracy for valence and arousal respectively. This database will significantly contribute to research on emotion recognition in the wild, while also providing a unique natural and freely accessible resource for Spanish., Comment: This paper has been superseded by arXiv:2403.02167 (merged from the description of the EMOVOME database in arXiv:2402.17496v1 and the speech emotion recognition models in arXiv:2403.02167v1)

There are different proposals in the literature on how to protect pedestrians using warning systems to alert drivers of their presence. They can be based on onboard perception systems or wireless communications. The evaluation of these systems has been focused on testing their ability to detect pedestrians. A problem that has received much less attention is the possibility of generating too many alerts in the warning systems. In this paper, we propose and analyze four different algorithms to take the decision on generating alerts in a warning system that is based on direct wireless communications between vehicles and pedestrians. With the algorithms, we explore different strategies to reduce unnecessary alerts. The feasibility of the implementation of the algorithms was evaluated with a deployment using real equipment, and tests were carried out to verify their behavior in real scenarios. The ability of each algorithm to reduce unnecessary alerts was evaluated with realistic simulations in an urban scenario, using a traffic simulator with vehicular and pedestrian flows. The results show the importance of tackling the problem of driver overload in warning systems, and that it is not straightforward to predict the load of alerts generated by an algorithm in a large-scale deployment, in which there are multiple interactions between vehicles and pedestrians.