7 results on '"Balanya, Sergio A."'
Search Results
2. Adaptive Temperature Scaling for Robust Calibration of Deep Neural Networks
- Author
-
Balanya, Sergio A., Maroñas, Juan, and Ramos, Daniel
- Subjects
Computer Science - Machine Learning - Abstract
In this paper, we study the post-hoc calibration of modern neural networks, a problem that has drawn a lot of attention in recent years. Many calibration methods of varying complexity have been proposed for the task, but there is no consensus about how expressive these should be. We focus on the task of confidence scaling, specifically on post-hoc methods that generalize Temperature Scaling, we call these the Adaptive Temperature Scaling family. We analyse expressive functions that improve calibration and propose interpretable methods. We show that when there is plenty of data complex models like neural networks yield better performance, but are prone to fail when the amount of data is limited, a common situation in certain post-hoc calibration applications like medical diagnosis. We study the functions that expressive methods learn under ideal conditions and design simpler methods but with a strong inductive bias towards these well-performing functions. Concretely, we propose Entropy-based Temperature Scaling, a simple method that scales the confidence of a prediction according to its entropy. Results show that our method obtains state-of-the-art performance when compared to others and, unlike complex models, it is robust against data scarcity. Moreover, our proposed model enables a deeper interpretation of the calibration process., Comment: 21 pages, journal article, 5 figures
- Published
- 2022
3. Gaussianization of LA-ICP-MS features to improve calibration in forensic glass comparison
- Author
-
Ramirez-Hereza, Pablo, Ramos, Daniel, Maroñas, Juan, Balanya, Sergio A., and Almirall, Jose
- Published
- 2023
- Full Text
- View/download PDF
4. Gaussian Processes for radiation dose prediction in nuclear power plant reactors
- Author
-
Balanya, Sergio A., Ramos, Daniel, Ramirez-Hereza, Pablo, Toledano, Doroteo T., Gonzalez-Rodriguez, Joaquin, Ariza-Velazquez, Alicia, Vidal Orlovac, Josip, and Doncel Gutiérrez, Nuria
- Published
- 2022
- Full Text
- View/download PDF
5. Gaussianization of LA-ICP-MS Features to Improve Calibration in Forensic Glass Comparison
- Author
-
Ramirez Hereza, Pablo, Ramos Castro, Daniel, Maroñas, Juan, Balanya, Sergio A., Almirall, José, and UAM. Departamento de Tecnología Electrónica y de las Comunicaciones
- Subjects
Telecomunicaciones ,Likelihood Ratio ,Forensic Glass Comparison ,Normalization flows ,Gaussianization ,Electrónica ,LA-ICP-MS ,Law ,Bayesian models ,Pathology and Forensic Medicine - Abstract
The forensic comparison of glass aims to compare a glass sample of an unknown source with a control glass sample of a known source. In this work, we use multi-elemental features from Laser Ablation Inductively Coupled Plasma with Mass Spectrometry (LA-ICP-MS) to compute a likelihood ratio. This calculation is a complex procedure that generally requires a probabilistic model including the within-source and betweensource variabilities of the features. Assuming the within-source variability to be normally distributed is a practical premise with the available data. However, the between-source variability is generally assumed to follow a much more complex distribution, typically described with a kernel density function. In this work, instead of modeling distributions with complex densities, we propose the use of simpler models and the introduction of a data pre-processing step consisting on the Gaussianization of the glass features. In this context, to obtain a better fit of the features with the Gaussian model assumptions, we explore the use of different normalization techniques of the LA-ICP-MS glass features, namely marginal Gaussianization based on histogram matching, marginal Gaussianization based on Yeo-Johnson transformation and a more complex joint Gaussianization using normalizing flows. We report an improvement in the performance of the Likelihood Ratios computed with the previously Gaussianized feature vectors, particularly relevant in their calibration, which implies a more reliable forensic glass comparison, This work has been supported by the Spanish Ministerio de Ciencia e Innovación through grant PID2021-125943OB-I00
- Published
- 2022
6. Técnicas expresivas de calibración para clasificadores multiclase
- Author
-
Alvarez Balanya, Sergio, Ramos, Daniel, Maroñas Molano, Juan, UAM. Departamento de Tecnología Electrónica y de las Comunicaciones, Ramos Castro, Daniel, Ramos, Daniel (director), and Maroñas Molano, Juan (director)
- Subjects
Deep Neural Networks ,Telecomunicaciones ,Calibration ,Bayesian Neural Networks - Abstract
Máster Universitario en en Investigación e Innovación en Inteligencia Computacional y Sistemas Interactivos, In the last years Deep Neural Networks (DNNs) have gained popularity as classi cation systems and are now applied to multiple tasks where they obtain state-of-the-art accuracy. Some of this tasks bene t from con dence scores associated to predictions. To this aim, a probabilistic classi er outputs con dence scores which are identi ed with the categorical distribution over the set of possible classes. In this context, calibration is identi ed with the validity of con dence scores as true probabilities. Although DNNs report state-of-the-art accuracy in a plethora of tasks, it has been pointed out recently that they present poor calibration. This has motivated a growing body of literature on the calibration of DNNs. One common approach, and the taken in this work, is that of training a separate model to map uncalibrated outputs to better calibrated predictions, namely re-calibration. We extend the work in [1] where the authors propose to use Bayesian Neural Networks (BNNs) as a decoupled stage from the main training of a DNN. The Bayesian approach is appealing in the sense that it presents proper uncertainty modeling. However, recent work [2] argues that current approximations to the standard Bayesian approach fail to recognize fundamental assumptions of the Bayesian paradigm. We apply the novel framework of Generalized Variational Inference (GVI) to approximate the posterior of the decoupled BNNs. In particular, we substitute the traditional KL divergence used in the standard Variational Inference method for a robust divergence. We provide a comprehensive comparison between both divergences. Moreover, since there is no much literature on the use of robust divergences along GVI, we also contribute with a sensitivity study in which we analyze the in uence of the divergence parameters
- Published
- 2020
7. Generación automática de conjuntos de evaluación de camuflaje
- Author
-
Álvarez Balanya, Sergio, Escudero-Viñolo, Marcos, Bescós, Jesús, and UAM. Departamento de Tecnología Electrónica y de las Comunicaciones
- Subjects
Telecomunicaciones ,camouflage ,Jung’s theorem ,background subtraction - Abstract
Background subtraction has become a key step in several computer vision algorithms. There are plenty of studies proposing different and varied approaches. However, the problem of background subtraction is not yet fully addressed. One reason might be the fact that each method has been developed for different tasks, e.g. video surveillance or optical motion capture. The recent appearance of comprehensive datasets provides a common framework for evaluating background subtraction algorithms. These datasets present a balanced repertoire of sequences in which common challenges are present. This leads to extensive overall scores in which robustness against different challenges is considered, but not particularized to these challenges. A particularly barely studied challenge, and the focus of our work, is camouflage: the resemblance between background and foreground samples. The research community agrees that there isn’t yet a commonly accepted approach to handle camouflage. In this work, we propose a novel solution for modeling camouflage based on the Jung’s theorem. Based on this solution, we generate camouflage likelihoods for every foreground pixel in a sequence using available ground-truth information to discriminate the background from the foreground. The evaluation of the proposed solution is performed in discrepancy terms by thresholding the camouflage likelihoods to obtain a binary mask on which we apply classical classification metrics. Thereby, we are able to further analyze the effect of the features selected by different background subtraction algorithms in handling camouflage. Furthermore, the proposed solution also permits the ranking of a set of sequences in terms of camouflage. The experiments carried out on the popular CDNET2014 dataset suggest that the use of certain alternative features to color—e,g, motion—is beneficial to robustly handle camouflage.
- Published
- 2018
Catalog
Discovery Service for Jio Institute Digital Library
For full access to our library's resources, please sign in.