A5026967773- Neural Networks and Applications
- Face and Expression Recognition
- Machine Learning and Data Classification
- Machine Learning and Algorithms
- Time Series Analysis and Forecasting
- Adversarial Robustness in Machine Learning
- Obsessive-Compulsive Spectrum Disorders
- Gaussian Processes and Bayesian Inference
- Fault Detection and Control Systems
- High voltage insulation and dielectric phenomena
- Advanced Wireless Communication Techniques
- Domain Adaptation and Few-Shot Learning
- Advanced Data Compression Techniques
- Machine Learning and ELM
- Autism Spectrum Disorder Research
- Power Transformer Diagnostics and Insulation
- Biochemical and Structural Characterization
- Satellite Communication Systems
- Blind Source Separation Techniques
- Music and Audio Processing
- Wireless Communication Networks Research
- Speech Recognition and Synthesis
- Speech and Audio Processing
- Music Technology and Sound Studies
- Fractal and DNA sequence analysis
Universidad Carlos III de Madrid
2012-2024
The fields of machine learning and mathematical programming are increasingly intertwined. Optimization problems lie at the heart most approaches. Special Topic on Machine Learning Large Scale examines this interplay. researchers have embraced advances in allowing new types models to be pursued. special topic includes using quadratic, linear, second-order cone, semi-definite, semi-infinite programs. We observe that qualities good optimization algorithms from perspectives can quite different....
The measurement of partial discharge (PD) signals in the radio frequency (RF) range has gained popularity among utilities and specialized monitoring companies recent years. Unfortunately, most occasions data are hidden by noise coupled interferences that hinder their interpretation renders them useless especially acquisition systems ultra high (UHF) band where interest weak. This paper is focused on a method uses selective spectral signal characterization to feature each signal, type or...
An important problem that hinders the use of supervised classification algorithms for brain imaging is number variables per single subject far exceeds training subjects available. Deriving multivariate measures variable importance becomes a challenge in such scenarios. This paper proposes new measure termed sign-consistency bagging (SCB). The SCB captures by analyzing sign consistency corresponding weights an ensemble linear support vector machine (SVM) classifiers. Further, importances are...
The measurement of the emitted electromagnetic energy in UHF region spectrum allows detection partial discharges and, thus, on-line monitoring condition insulation electrical equipment. Unfortunately, determining affected asset is difficult when there are several simultaneous defects. This paper proposes use an independent component analysis (ICA) algorithm to separate signals coming from different discharge (PD) sources. performance has been tested using generated by test objects. results...
This paper presents a system for the detection of partial discharges (PD) in industrial applications based on One Class Support Vector Machines (OCSVM). The study stresses Partial Discharges as they represent major source information related to degradation equipment. PD measurement is widely extended technique condition monitoring electrical machines and power cables avoid catastrophic failures consequent blackouts. most important keystones interpretation their separation from other signals...
Adaptive learning is necessary for nonstationary environments where the machine needs to forget past data distribution. Efficient algorithms require a compact model update not grow in computational burden with incoming and lowest possible cost online parameter updating. Existing solutions only partially cover these needs. Here, we propose first adaptive sparse Gaussian process (GP) able address all issues. We reformulate variational GP (VSGP) algorithm make it through forgetting factor....
Concept-bottleneck models (CBMs) are a new paradigm to construct interpretable classifiers. The CBM architecture can be regarded as neural network with single hidden layer whose neurons replaced by binary Each of these classifiers implements concept, that is, it attempts answer meaningful yes/no question: the presence or absence concept in observation presented network. output is usually implemented linear stage combines concepts into final decisions. This paper presents an application...
This paper introduces a novel approach to learn multi-task regression models with constrained architecture complexity. The proposed model, named RFF-BLR, consists of randomised feedforward neural network two fundamental characteristics: single hidden layer whose units implement the random Fourier features that approximate an RBF kernel, and Bayesian formulation optimises weights connecting output layers. RFF-based inherits robustness kernel methods. enables promoting multioutput sparsity:...
This paper evaluates the capabilities of model-based distances between time series to identify musical genre songs. In contrast with standard approaches, this kind metrics can take into account structure songs by modeling dynamics parameter sequences. We tackle problem from a non-supervised and supervised perspective, in order point out usefulness dynamic-based distances. Experiments on real-world dataset containing genres different degrees priori overlapping give insights about discriminant these
PAC-Bayes bounds have been proposed to get risk estimates based on a training sample. In this paper the approach is combined with stability of hypothesis learned by Hilbert space valued algorithm. The setting used Gaussian prior centered at expected output. Thus novelty our using priors defined in terms data-generating distribution. Our main result randomized algorithm coefficients. We also provide new bound for SVM classifier, which compared other known experimentally. Ours appears be first...