Browsing by Author "Carrasco, Miguel"

Now showing 1 - 7 of 7
  • Thumbnail Image
    Item
    33
    (33, 33) Carrasco, Miguel; Vairetti, Carla; López, Julio; Maldonado, Sebastián; 33
    Kernel methods are crucial in machine learning due to their ability to model nonlinear relationships in data. Among these, Support Vector Machine (SVM) is widely recognized for its robust performance and appealing optimization properties. In this work, we build upon recent advancements in SVM variants to propose five novel models specifically designed for multiclass learning. In particular, we introduce One-vs-One and One-vs-All versions of the nonparallel hyperplane SVM and improved twin SVM, along with a unified optimization variant (all-together) of the former method for nonlinear multiclass classification. Our empirical evaluation, conducted on 11 datasets and 12 multiclass classifiers, shows the superiority of our methods: four out of the five proposed models rank among the top performers and consistently outperform alternative approaches in terms of balanced accuracy. Additionally, a statistical test was performed, showing significant differences among the classifiers.
  • Thumbnail Image
    Item
    33
    (33, 33) Canelas, Alfredo; Carrasco, Miguel; López, Julio; Paduro, Esteban; 33
    This article introduces an open-source MATLAB package that implements an interior point algorithm for numerically finding the minimum of a Nonlinear Second-Order Cone Program (NSOCP). The implementation is based on the Feasible Direction Interior Point Algorithm for NSOCP (FDIPA-SOC). We have designed a user-friendly interface similar to MATLAB’s fmincon routine, and we will demonstrate how to utilize it by solving several examples of NSOCP found in the literature. Finally, we showcase an application to Machine Learning (Robust Support Vector Machine) using the software. To the best of our knowledge, this is the first open-source software made available to the scientific community for solving NSOCP.
  • Thumbnail Image
    Item
    33
    (33, 33) Cisternas, Jaime; Mellado, Paula; Urbina, Felipe; Portilla, Cristóbal; Carrasco, Miguel; Concha, Andrés; 33
    In classical mechanics, solutions can be classified according to their stability. Each of them is part of the possible trajectories of the system. However, the signatures of unstable solutions are hard to observe in an experiment, and most of the times if the experimental realization is adiabatic, they are considered just a nuisance. Here we use a small number of XY magnetic dipoles subject to an external magnetic field for studying the origin of their collective magnetic response. Using bifurcation theory we have found all the possible solutions being stable or unstable, and explored how those solutions are naturally connected by points where the symmetries of the system are lost or restored. Unstable solutions that reveal the symmetries of the system are found to be the culprit that shape hysteresis loops in this system. The complexity of the solutions for the nonlinear dynamics is analyzed using the concept of boundary basin entropy, finding that the damping timescale is critical for the emergence of fractal structures in the basins of attraction. Furthermore, we numerically found domain wall solutions that are the smallest possible realizations of transverse walls and vortex walls in magnetism. We experimentally confirmed their existence and stability showing that our system is a suitable platform to study domain wall dynamics at the macroscale.
  • Thumbnail Image
    Item
    33
    (33, 33) Maldonado, Sebastián; Vairetti, Carla; Jara, Katherine; Carrasco, Miguel; López, Julio; 33
    In this paper, we propose a novel adaptive loss function for enhancing deep learning performance in classification tasks. Specifically, we redefine the cross-entropy loss to effectively address class-level noise conditions, including the challenging problem of class imbalance. Our approach introduces aggregation operators to improve classification accuracy. The rationale behind our proposed method lies in the iterative up-weighting of class-level components within the loss function, focusing on those with larger errors. To achieve this, we employ the ordered weighted average (OWA) operator and combine it with an adaptive scheme for gradient-based learning. The main finding is that our method outperforms other commonly used loss functions, such as the standard cross-entropy or focal loss, across various binary and multiclass classification tasks. Furthermore, we explore the influence of hyperparameters associated with the OWA operators and propose a default configuration that performs well across different experimental settings.
  • Thumbnail Image
    Item
    33
    (33, 33) Carrasco, Miguel; Ivorra, Benjamin; López, Julio; Ramos, Angel M.; 33
    Methods: Feature selection is essential for building effective machine learning models in binary classification. Eliminating unnecessary features can reduce the risk of overfitting and improve classification performance. Moreover, the data we handle typically contains a stochastic component, making it important to develop robust models that are insensitive to data perturbations. Although there are numerous methods and tools for feature selection, relatively few studies address embedded feature selection within robust classification models using penalization techniques. Objective: In this work, we introduce robust classifiers with integrated feature selection capabilities, utilizing probability machines based on different penalization techniques, such as the ?1-norm or the elastic-net, combined with a novel Direct Feature Elimination process to improve model resilience and efficiency. Findings: Numerical experiments on standard datasets demonstrate the effectiveness and robustness of the proposed models in classification tasks even when using a reduced number of features. These experiments were evaluated using original performance indicators, highlighting the models’ ability to maintain high performance with fewer features. Novelty: The study discusses the trade-offs involved in combining different penalties to select the most relevant features while minimizing empirical risk. In particular, the integration of elastic-net and ?1-norm penalties within a unified framework, combined with the original Direct Feature Elimination approach, presents a novel method for improving both model accuracy and robustness.
  • Thumbnail Image
    Item
    The Cobb-Douglas Learning Machine
    Maldonado, Sebastián; López, Julio; Carrasco, Miguel
  • Thumbnail Image
    Item