Pablo M. Olmos

Batuhan Koyuncu, Pablo Sanchez-Martin, Ignacio Peis et al.

Recent approaches build on implicit neural representations (INRs) to propose generative models over function spaces. However, they are computationally costly when dealing with inference tasks, such as missing data imputation, or directly cannot tackle them. In this work, we propose a novel deep generative model, named VAMoH. VAMoH combines the capabilities of modeling continuous functions using INRs and the inference capabilities of Variational Autoencoders (VAEs). In addition, VAMoH relies on a normalizing flow to define the prior, and a mixture of hypernetworks to parametrize the data log-likelihood. This gives VAMoH a high expressive capability and interpretability. Through experiments on a diverse range of data types, such as images, voxels, and climate data, we show that VAMoH can effectively learn rich distributions over continuous functions. Furthermore, it can perform inference-related tasks, such as conditional super-resolution generation and in-painting, as well or better than previous approaches, while being less computationally demanding.

Antía López Galdo, Alejandro Guerrero-López, Pablo M. Olmos et al.

Railway axle maintenance is critical to avoid catastrophic failures. Nowadays, condition monitoring techniques are becoming more prominent in the industry to prevent enormous costs and damage to human lives. This paper proposes the development of a railway axle condition monitoring system based on advanced 2D-Convolutional Neural Network (CNN) architectures applied to time-frequency representations of vibration signals. For this purpose, several preprocessing steps and different types of Deep Learning (DL) and Machine Learning (ML) architectures are discussed to design an accurate classification system. The resultant system converts the railway axle vibration signals into time-frequency domain representations, i.e., spectrograms, and, thus, trains a two-dimensional CNN to classify them depending on their cracks. The results showed that the proposed approach outperforms several alternative methods tested. The CNN architecture has been tested in 3 different wheelset assemblies, achieving AUC scores of 0.93, 0.86, and 0.75 outperforming any other architecture and showing a high level of reliability when classifying 4 different levels of defects.

María Martínez-García, Fernando Moreno-Pino, Pablo M. Olmos et al.

Sleep constitutes a key indicator of human health, performance, and quality of life. Sleep deprivation has long been related to the onset, development, and worsening of several mental and metabolic disorders, constituting an essential marker for preventing, evaluating, and treating different health conditions. Sleep Activity Recognition methods can provide indicators to assess, monitor, and characterize subjects' sleep-wake cycles and detect behavioral changes. In this work, we propose a general method that continuously operates on passively sensed data from smartphones to characterize sleep and identify significant sleep episodes. Thanks to their ubiquity, these devices constitute an excellent alternative data source to profile subjects' biorhythms in a continuous, objective, and non-invasive manner, in contrast to traditional sleep assessment methods that usually rely on intrusive and subjective procedures. A Heterogeneous Hidden Markov Model is used to model a discrete latent variable process associated with the Sleep Activity Recognition task in a self-supervised way. We validate our results against sleep metrics reported by tested wearables, proving the effectiveness of the proposed approach and advocating its use to assess sleep without more reliable sources.

Fernando Moreno-Pino, María Martínez-García, Pablo M. Olmos et al.

Psychiatric patients' passive activity monitoring is crucial to detect behavioural shifts in real-time, comprising a tool that helps clinicians supervise patients' evolution over time and enhance the associated treatments' outcomes. Frequently, sleep disturbances and mental health deterioration are closely related, as mental health condition worsening regularly entails shifts in the patients' circadian rhythms. Therefore, Sleep Activity Recognition constitutes a behavioural marker to portray patients' activity cycles and to detect behavioural changes among them. Moreover, mobile passively sensed data captured from smartphones, thanks to these devices' ubiquity, constitute an excellent alternative to profile patients' biorhythm. In this work, we aim to identify major sleep episodes based on passively sensed data. To do so, a Heterogeneous Hidden Markov Model is proposed to model a discrete latent variable process associated with the Sleep Activity Recognition task in a self-supervised way. We validate our results against sleep metrics reported by clinically tested wearables, proving the effectiveness of the proposed approach.

Óscar Jiménez Rama, Fernando Moreno-Pino, David Ramírez et al.

The best encoding is the one that is interpretable in nature. In this work, we introduce a novel model that incorporates an interpretable bottleneck-termed the Filter Bank (FB)-at the outset of a Variational Autoencoder (VAE). This arrangement compels the VAE to attend on the most informative segments of the input signal, fostering the learning of a novel encoding ${f_0}$ which boasts enhanced interpretability and clusterability over traditional latent spaces. By deliberately constraining the VAE with this FB, we intentionally constrict its capacity to access broad input domain information, promoting the development of an encoding that is discernible, separable, and of reduced dimensionality. The evolutionary learning trajectory of ${f_0}$ further manifests as a dynamic hierarchical tree, offering profound insights into cluster similarities. Additionally, for handling intricate data configurations, we propose a tailored decoder structure that is symmetrically aligned with FB's architecture. Empirical evaluations highlight the superior efficacy of ISVAE, which compares favorably to state-of-the-art results in clustering metrics across real-world datasets.

Alejandro Guerrero-López, Carlos Sevilla-Salcedo, Vanessa Gómez-Verdejo et al.

Purpose: Handling heterogeneous and mixed data types has become increasingly critical with the exponential growth in real-world databases. While deep generative models attempt to merge diverse data views into a common latent space, they often sacrifice interpretability, flexibility, and modularity. This study proposes a novel method to address these limitations by combining Variational AutoEncoders (VAEs) with a Factor Analysis latent space (FA-VAE). Methods: The proposed FA-VAE method employs multiple VAEs to learn a private representation for each heterogeneous data view in a continuous latent space. Information is shared between views using a low-dimensional latent space, generated via a linear projection matrix. This modular design creates a hierarchical dependency between private and shared latent spaces, allowing for the flexible addition of new views and conditioning of pre-trained models. Results: The FA-VAE approach facilitates cross-generation of data from different domains and enables transfer learning between generative models. This allows for effective integration of information across diverse data views while preserving their distinct characteristics. Conclusions: By overcoming the limitations of existing methods, the FA-VAE provides a more interpretable, flexible, and modular solution for managing heterogeneous data types. It offers a pathway to more efficient and scalable data-handling strategies, enhancing the potential for cross-domain data synthesis and model transferability.

Elias Abad Rocamora, Fanghui Liu, Grigorios G. Chrysos et al.

Catastrophic overfitting (CO) in single-step adversarial training (AT) results in abrupt drops in the adversarial test accuracy (even down to 0%). For models trained with multi-step AT, it has been observed that the loss function behaves locally linearly with respect to the input, this is however lost in single-step AT. To address CO in single-step AT, several methods have been proposed to enforce local linearity of the loss via regularization. However, these regularization terms considerably slow down training due to Double Backpropagation. Instead, in this work, we introduce a regularization term, called ELLE, to mitigate CO effectively and efficiently in classical AT evaluations, as well as some more difficult regimes, e.g., large adversarial perturbations and long training schedules. Our regularization term can be theoretically linked to curvature of the loss function and is computationally cheaper than previous methods by avoiding Double Backpropagation. Our thorough experimental validation demonstrates that our work does not suffer from CO, even in challenging settings where previous works suffer from it. We also notice that adapting our regularization parameter during training (ELLE-A) greatly improves the performance, specially in large $ε$ setups. Our implementation is available in https://github.com/LIONS-EPFL/ELLE .

Fernando Moreno-Pino, Emese Sükei, Pablo M. Olmos et al.

We introduce PyHHMM, an object-oriented open-source Python implementation of Heterogeneous-Hidden Markov Models (HHMMs). In addition to HMM's basic core functionalities, such as different initialization algorithms and classical observations models, i.e., continuous and multinoulli, PyHHMM distinctively emphasizes features not supported in similar available frameworks: a heterogeneous observation model, missing data inference, different model order selection criterias, and semi-supervised training. These characteristics result in a feature-rich implementation for researchers working with sequential data. PyHHMM relies on the numpy, scipy, scikit-learn, and seaborn Python packages, and is distributed under the Apache-2.0 License. PyHHMM's source code is publicly available on Github (https://github.com/fmorenopino/HeterogeneousHMM) to facilitate adoptions and future contributions. A detailed documentation (https://pyhhmm.readthedocs.io/en/latest), which covers examples of use and models' theoretical explanation, is available. The package can be installed through the Python Package Index (PyPI), via 'pip install pyhhmm'.

Ignacio Peis, Pablo M. Olmos, Jes Frellsen

Order-Agnostic autoregressive models have demonstrated strong performance in deep generative modeling, yet their use in settings with incomplete data remains largely unexplored. In this work, we reinterpret them through the lens of missing data. First, we show that their standard training procedure on fully observed data implicitly performs imputation under a missing completely at random mechanism, resulting in robust out-of-sample imputation performance in settings with high missingness. Second, we introduce the first principled framework for training them directly on incomplete datasets under general missingness mechanisms. Third, we leverage their amortized conditional density estimation to perform active information acquisition, i.e., sequentially selecting the most informative missing variables for downstream prediction or inference. Across a suite of real-world benchmarks, our Missingness-Aware Order-Agnostic Autoregressive Model (MO-ARM) consistently outperforms established imputation baselines.

Mikkel Jordahn, Pablo M. Olmos

Deep Neural Networks (DNN) have shown great promise in many classification applications, yet are widely known to have poorly calibrated predictions when they are over-parametrized. Improving DNN calibration without comprising on model accuracy is of extreme importance and interest in safety critical applications such as in the health-care sector. In this work, we show that decoupling the training of feature extraction layers and classification layers in over-parametrized DNN architectures such as Wide Residual Networks (WRN) and Visual Transformers (ViT) significantly improves model calibration whilst retaining accuracy, and at a low training cost. In addition, we show that placing a Gaussian prior on the last hidden layer outputs of a DNN, and training the model variationally in the classification training stage, even further improves calibration. We illustrate these methods improve calibration across ViT and WRN architectures for several image classification benchmark datasets.

Lorena Gallego-Viñarás, Juan Miguel Mira-Tomás, Anna Michela-Gaeta et al.

Alzheimer's disease (AD) and sleep disorders exhibit a close association, where disruptions in sleep patterns often precede the onset of Mild Cognitive Impairment (MCI) and early-stage AD. This study delves into the potential of utilizing sleep-related electroencephalography (EEG) signals acquired through polysomnography (PSG) for the early detection of AD. Our primary focus is on exploring semi-supervised Deep Learning techniques for the classification of EEG signals due to the clinical scenario characterized by the limited data availability. The methodology entails testing and comparing the performance of semi-supervised SMATE and TapNet models, benchmarked against the supervised XCM model, and unsupervised Hidden Markov Models (HMMs). The study highlights the significance of spatial and temporal analysis capabilities, conducting independent analyses of each sleep stage. Results demonstrate the effectiveness of SMATE in leveraging limited labeled data, achieving stable metrics across all sleep stages, and reaching 90% accuracy in its supervised form. Comparative analyses reveal SMATE's superior performance over TapNet and HMM, while XCM excels in supervised scenarios with an accuracy range of 92 - 94%. These findings underscore the potential of semi-supervised models in early AD detection, particularly in overcoming the challenges associated with the scarcity of labeled data. Ablation tests affirm the critical role of spatio-temporal feature extraction in semi-supervised predictive performance, and t-SNE visualizations validate the model's proficiency in distinguishing AD patterns. Overall, this research contributes to the advancement of AD detection through innovative Deep Learning approaches, highlighting the crucial role of semi-supervised learning in addressing data limitations.

José Manuel de Frutos, Pablo M. Olmos, Manuel A. Vázquez et al.

Implicit generative models have the capability to learn arbitrary complex data distributions. On the downside, training requires telling apart real data from artificially-generated ones using adversarial discriminators, leading to unstable training and mode-dropping issues. As reported by Zahee et al. (2017), even in the one-dimensional (1D) case, training a generative adversarial network (GAN) is challenging and often suboptimal. In this work, we develop a discriminator-free method for training one-dimensional (1D) generative implicit models and subsequently expand this method to accommodate multivariate cases. Our loss function is a discrepancy measure between a suitably chosen transformation of the model samples and a uniform distribution; hence, it is invariant with respect to the true distribution of the data. We first formulate our method for 1D random variables, providing an effective solution for approximate reparameterization of arbitrary complex distributions. Then, we consider the temporal setting (both univariate and multivariate), in which we model the conditional distribution of each sample given the history of the process. We demonstrate through numerical simulations that this new method yields promising results, successfully learning true distributions in a variety of scenarios and mitigating some of the well-known problems that state-of-the-art implicit methods present.

Darío C. Larese, Almudena Bravo Cerrada, Gabriel Dambrosio Tomei et al.

Maintaining railway axles is critical to preventing severe accidents and financial losses. The railway industry is increasingly interested in advanced condition monitoring techniques to enhance safety and efficiency, moving beyond traditional periodic inspections toward Maintenance 4.0. This study introduces a robust Deep Autoregressive solution that integrates seamlessly with existing systems to avert mechanical failures. Our approach simulates and predicts vibration signals under various conditions and fault scenarios, improving dataset robustness for more effective detection systems. These systems can alert maintenance needs, preventing accidents preemptively. We use experimental vibration signals from accelerometers on train axles. Our primary contributions include a transformer model, ShaftFormer, designed for processing time series data, and an alternative model incorporating spectral methods and enhanced observation models. Simulating vibration signals under diverse conditions mitigates the high cost of obtaining experimental signals for all scenarios. Given the non-stationary nature of railway vibration signals, influenced by speed and load changes, our models address these complexities, offering a powerful tool for predictive maintenance in the rail industry.

Juan José Murillo-Fuentes, Pablo M. Olmos, Laura Alba-Carcelén

The study of canvas fabrics in works of art is a crucial tool for authentication, attribution and conservation. Traditional methods are based on thread density map matching, which cannot be applied when canvases do not come from contiguous positions on a roll. This paper presents a novel approach based on deep learning to assess the similarity of textiles. We introduce an automatic tool that evaluates the similarity between canvases without relying on thread density maps. A Siamese deep learning model is designed and trained to compare pairs of images by exploiting the feature representations learned from the scans. In addition, a similarity estimation method is proposed, aggregating predictions from multiple pairs of cloth samples to provide a robust similarity score. Our approach is applied to canvases from the Museo Nacional del Prado, corroborating the hypothesis that plain weave canvases, widely used in painting, can be effectively compared even when their thread densities are similar. The results demonstrate the feasibility and accuracy of the proposed method, opening new avenues for the analysis of masterpieces.

José Manuel de Frutos, Manuel A. Vázquez, Pablo M. Olmos et al.

Rank-based statistical metrics, such as the invariant statistical loss (ISL), have recently emerged as robust and practically effective tools for training implicit generative models. In this work, we introduce dual-ISL, a novel likelihood-free objective for training implicit generative models that interchanges the roles of the target and model distributions in the ISL framework, yielding a convex optimization problem in the space of model densities. We prove that the resulting rank-based discrepancy $d_K$ is i) continuous under weak convergence and with respect to the $L^1$ norm, and ii) convex in its first argument-properties not shared by classical divergences such as KL or Wasserstein distances. Building on this, we develop a theoretical framework that interprets $d_K$ as an $L^2$-projection of the density ratio $q = p/\tilde p$ onto a Bernstein polynomial basis, from which we derive exact bounds on the truncation error, precise convergence rates, and a closed-form expression for the truncated density approximation. We further extend our analysis to the multivariate setting via random one-dimensional projections, defining a sliced dual-ISL divergence that retains both convexity and continuity. We empirically show that these theoretical advantages translate into practical ones. Specifically, across several benchmarks dual-ISL converges more rapidly, delivers markedly smoother and more stable training, and more effectively prevents mode collapse than classical ISL and other leading implicit generative methods-while also providing an explicit density approximation.

Rodrigo Oliver, Josué Pérez-Sabater, Leire Paz-Arbaizar et al.

Foundation models have achieved remarkable success across various domains, yet their adoption in healthcare remains limited. While significant advances have been made in medical imaging, genetic biomarkers, and time series from electronic health records, the potential of foundation models for patient behavior monitoring through personal digital devices remains underexplored. The data generated by these devices are inherently heterogeneous, multisource, and often exhibit high rates of missing data, posing unique challenges. This paper introduces a novel foundation model based on a modified vector quantized variational autoencoder, specifically designed to process real-world data from smartphones and wearable devices. We leveraged the discrete latent representation of this model to effectively perform two downstream tasks, suicide risk assessment and emotional state prediction, on different held-out clinical cohorts without the need of fine-tuning. We also highlight the existence of a trade-off between discrete and continuous latent structures, suggesting that hybrid models may be optimal for balancing accuracy across various supervised and unsupervised tasks.

Zi Yang, Ying Li, Zhidi Lin et al.

The multi-view Gaussian process latent variable model (MV-GPLVM) aims to learn a unified representation from multi-view data but is hindered by challenges such as limited kernel expressiveness and low computational efficiency. To overcome these issues, we first introduce a new duality between the spectral density and the kernel function. By modeling the spectral density with a bivariate Gaussian mixture, we then derive a generic and expressive kernel termed Next-Gen Spectral Mixture (NG-SM) for MV-GPLVMs. To address the inherent computational inefficiency of the NG-SM kernel, we propose a random Fourier feature approximation. Combined with a tailored reparameterization trick, this approximation enables scalable variational inference for both the model and the unified latent representations. Numerical evaluations across a diverse range of multi-view datasets demonstrate that our proposed method consistently outperforms state-of-the-art models in learning meaningful latent representations.

Ying Li, Zhidi Lin, Yuhao Liu et al.

Random feature latent variable models (RFLVMs) represent the state-of-the-art in latent variable models, capable of handling non-Gaussian likelihoods and effectively uncovering patterns in high-dimensional data. However, their heavy reliance on Monte Carlo sampling results in scalability issues which makes it difficult to use these models for datasets with a massive number of observations. To scale up RFLVMs, we turn to the optimization-based variational Bayesian inference (VBI) algorithm which is known for its scalability compared to sampling-based methods. However, implementing VBI for RFLVMs poses challenges, such as the lack of explicit probability distribution functions (PDFs) for the Dirichlet process (DP) in the kernel learning component, and the incompatibility of existing VBI algorithms with RFLVMs. To address these issues, we introduce a stick-breaking construction for DP to obtain an explicit PDF and a novel VBI algorithm called ``block coordinate descent variational inference" (BCD-VI). This enables the development of a scalable version of RFLVMs, or in short, SRFLVM. Our proposed method shows scalability, computational efficiency, superior performance in generating informative latent representations and the ability of imputing missing data across various real-world datasets, outperforming state-of-the-art competitors.

Carlos Sevilla-Salcedo, Vandad Imani, Pablo M. Olmos et al.

Machine learning techniques typically applied to dementia forecasting lack in their capabilities to jointly learn several tasks, handle time dependent heterogeneous data and missing values. In this paper, we propose a framework using the recently presented SSHIBA model for jointly learning different tasks on longitudinal data with missing values. The method uses Bayesian variational inference to impute missing values and combine information of several views. This way, we can combine different data-views from different time-points in a common latent space and learn the relations between each time-point while simultaneously modelling and predicting several output variables. We apply this model to predict together diagnosis, ventricle volume, and clinical scores in dementia. The results demonstrate that SSHIBA is capable of learning a good imputation of the missing values and outperforming the baselines while simultaneously predicting three different tasks.

Fernando Moreno-Pino, Pablo M. Olmos, Antonio Artés-Rodríguez

Time series forecasting is an important problem across many domains, playing a crucial role in multiple real-world applications. In this paper, we propose a forecasting architecture that combines deep autoregressive models with a Spectral Attention (SA) module, which merges global and local frequency domain information in the model's embedded space. By characterizing in the spectral domain the embedding of the time series as occurrences of a random process, our method can identify global trends and seasonality patterns. Two spectral attention models, global and local to the time series, integrate this information within the forecast and perform spectral filtering to remove time series's noise. The proposed architecture has a number of useful properties: it can be effectively incorporated into well-know forecast architectures, requiring a low number of parameters and producing interpretable results that improve forecasting accuracy. We test the Spectral Attention Autoregressive Model (SAAM) on several well-know forecast datasets, consistently demonstrating that our model compares favorably to state-of-the-art approaches.

Daniel Barrejón, Pablo M. Olmos, Antonio Artés-Rodríguez

Medical data sets are usually corrupted by noise and missing data. These missing patterns are commonly assumed to be completely random, but in medical scenarios, the reality is that these patterns occur in bursts due to sensors that are off for some time or data collected in a misaligned uneven fashion, among other causes. This paper proposes to model medical data records with heterogeneous data types and bursty missing data using sequential variational autoencoders (VAEs). In particular, we propose a new methodology, the Shi-VAE, which extends the capabilities of VAEs to sequential streams of data with missing observations. We compare our model against state-of-the-art solutions in an intensive care unit database (ICU) and a dataset of passive human monitoring. Furthermore, we find that standard error metrics such as RMSE are not conclusive enough to assess temporal models and include in our analysis the cross-correlation between the ground truth and the imputed signal. We show that Shi-VAE achieves the best performance in terms of using both metrics, with lower computational complexity than the GP-VAE model, which is the state-of-the-art method for medical records.

Ignacio Peis, Pablo M. Olmos, Antonio Artés-Rodríguez

We present a novel deep generative model based on non i.i.d. variational autoencoders that captures global dependencies among observations in a fully unsupervised fashion. In contrast to the recent semi-supervised alternatives for global modeling in deep generative models, our approach combines a mixture model in the local or data-dependent space and a global Gaussian latent variable, which lead us to obtain three particular insights. First, the induced latent global space captures interpretable disentangled representations with no user-defined regularization in the evidence lower bound (as in $β$-VAE and its generalizations). Second, we show that the model performs domain alignment to find correlations and interpolate between different databases. Finally, we study the ability of the global space to discriminate between groups of observations with non-trivial underlying structures, such as face images with shared attributes or defined sequences of digits images.

José Carlos Aradillas, Juan José Murillo-Fuentes, Pablo M. Olmos

In this paper, we face the problem of offline handwritten text recognition (HTR) in historical documents when few labeled samples are available and some of them contain errors in the train set. Three main contributions are developed. First we analyze how to perform transfer learning (TL) from a massive database to a smaller historical database, analyzing which layers of the model need a fine-tuning process. Second, we analyze methods to efficiently combine TL and data augmentation (DA). Finally, an algorithm to mitigate the effects of incorrect labelings in the training set is proposed. The methods are analyzed over the ICFHR 2018 competition database, Washington and Parzival. Combining all these techniques, we demonstrate a remarkable reduction of CER (up to 6% in some cases) in the test set with little complexity overhead.

Carlos Sevilla-Salcedo, Alejandro Guerrero-López, Pablo M. Olmos et al.

Multi-view problems can be faced with latent variable models since they are able to find low-dimensional projections that fairly capture the correlations among the multiple views that characterise each datum. On the other hand, high-dimensionality and non-linear issues are traditionally handled by kernel methods, inducing a (non)-linear function between the latent projection and the data itself. However, they usually come with scalability issues and exposition to overfitting. Here, we propose merging both approaches into single model so that we can exploit the best features of multi-view latent models and kernel methods and, moreover, overcome their limitations. In particular, we combine probabilistic factor analysis with what we refer to as kernelized observations, in which the model focuses on reconstructing not the data itself, but its relationship with other data points measured by a kernel function. This model can combine several types of views (kernelized or not), and it can handle heterogeneous data and work in semi-supervised settings. Additionally, by including adequate priors, it can provide compact solutions for the kernelized observations -- based in a automatic selection of Bayesian Relevance Vectors (RVs) -- and can include feature selection capabilities. Using several public databases, we demonstrate the potential of our approach (and its extensions) w.r.t. common multi-view learning models such as kernel canonical correlation analysis or manifold relevance determination.

Carlos Sevilla-Salcedo, Vanessa Gómez-Verdejo, Pablo M. Olmos

The Bayesian approach to feature extraction, known as factor analysis (FA), has been widely studied in machine learning to obtain a latent representation of the data. An adequate selection of the probabilities and priors of these bayesian models allows the model to better adapt to the data nature (i.e. heterogeneity, sparsity), obtaining a more representative latent space. The objective of this article is to propose a general FA framework capable of modelling any problem. To do so, we start from the Bayesian Inter-Battery Factor Analysis (BIBFA) model, enhancing it with new functionalities to be able to work with heterogeneous data, include feature selection, and handle missing values as well as semi-supervised problems. The performance of the proposed model, Sparse Semi-supervised Heterogeneous Interbattery Bayesian Analysis (SSHIBA) has been tested on 4 different scenarios to evaluate each one of its novelties, showing not only a great versatility and an interpretability gain, but also outperforming most of the state-of-the-art algorithms.

Ignacio Peis, Pablo M. Olmos, Constanza Vera-Varela et al.

This article presents a novel method for predicting suicidal ideation from Electronic Health Records (EHR) and Ecological Momentary Assessment (EMA) data using deep sequential models. Both EHR longitudinal data and EMA question forms are defined by asynchronous, variable length, randomly-sampled data sequences. In our method, we model each of them with a Recurrent Neural Network (RNN), and both sequences are aligned by concatenating the hidden state of each of them using temporal marks. Furthermore, we incorporate attention schemes to improve performance in long sequences and time-independent pre-trained schemes to cope with very short sequences. Using a database of 1023 patients, our experimental results show that the addition of EMA records boosts the system recall to predict the suicidal ideation diagnosis from 48.13% obtained exclusively from EHR-based state-of-the-art methods to 67.78%. Additionally, our method provides interpretability through the t-SNE representation of the latent space. Further, the most relevant input features are identified and interpreted medically.

Pablo Sánchez-Martín, Pablo M. Olmos, Fernando Perez-Cruz

We propose a novel training procedure for improving the performance of generative adversarial networks (GANs), especially to bidirectional GANs. First, we enforce that the empirical distribution of the inverse inference network matches the prior distribution, which favors the generator network reproducibility on the seen samples. Second, we have found that the marginal log-likelihood of the samples shows a severe overrepresentation of a certain type of samples. To address this issue, we propose to train the bidirectional GAN using a non-uniform sampling for the mini-batch selection, resulting in improved quality and variety in generated samples measured quantitatively and by visual inspection. We illustrate our new procedure with the well-known CIFAR10, Fashion MNIST and CelebA datasets.

Fernando Perez-Cruz, Pablo M. Olmos, Michael Minyi Zhang et al.

In this paper, we take a new approach for time of arrival geo-localization. We show that the main sources of error in metropolitan areas are due to environmental imperfections that bias our solutions, and that we can rely on a probabilistic model to learn and compensate for them. The resulting localization error is validated using measurements from a live LTE cellular network to be less than 10 meters, representing an order-of-magnitude improvement.

Pablo Sánchez-Martín, Pablo M. Olmos, Fernando Pérez-Cruz

We propose a new method to evaluate GANs, namely EvalGAN. EvalGAN relies on a test set to directly measure the reconstruction quality in the original sample space (no auxiliary networks are necessary), and it also computes the (log)likelihood for the reconstructed samples in the test set. Further, EvalGAN is agnostic to the GAN algorithm and the dataset. We decided to test it on three state-of-the-art GANs over the well-known CIFAR-10 and CelebA datasets.

Alfredo Nazabal, Pablo M. Olmos, Zoubin Ghahramani et al.

Variational autoencoders (VAEs), as well as other generative models, have been shown to be efficient and accurate for capturing the latent structure of vast amounts of complex high-dimensional data. However, existing VAEs can still not directly handle data that are heterogenous (mixed continuous and discrete) or incomplete (with missing data at random), which is indeed common in real-world applications. In this paper, we propose a general framework to design VAEs suitable for fitting incomplete heterogenous data. The proposed HI-VAE includes likelihood models for real-valued, positive real valued, interval, categorical, ordinal and count data, and allows accurate estimation (and potentially imputation) of missing data. Furthermore, HI-VAE presents competitive predictive performance in supervised tasks, outperforming supervised models when trained on incomplete data.

José Carlos Aradillas, Juan José Murillo-Fuentes, Pablo M. Olmos

In this paper we deal with the offline handwriting text recognition (HTR) problem with reduced training datasets. Recent HTR solutions based on artificial neural networks exhibit remarkable solutions in referenced databases. These deep learning neural networks are composed of both convolutional (CNN) and long short-term memory recurrent units (LSTM). In addition, connectionist temporal classification (CTC) is the key to avoid segmentation at character level, greatly facilitating the labeling task. One of the main drawbacks of the CNNLSTM-CTC (CLC) solutions is that they need a considerable part of the text to be transcribed for every type of calligraphy, typically in the order of a few thousands of lines. Furthermore, in some scenarios the text to transcribe is not that long, e.g. in the Washington database. The CLC typically overfits for this reduced number of training samples. Our proposal is based on the transfer learning (TL) from the parameters learned with a bigger database. We first investigate, for a reduced and fixed number of training samples, 350 lines, how the learning from a large database, the IAM, can be transferred to the learning of the CLC of a reduced database, Washington. We focus on which layers of the network could be not re-trained. We conclude that the best solution is to re-train the whole CLC parameters initialized to the values obtained after the training of the CLC from the larger database. We also investigate results when the training size is further reduced. The differences in the CER are more remarkable when training with just 350 lines, a CER of 3.3% is achieved with TL while we have a CER of 18.2% when training from scratch. As a byproduct, the learning times are quite reduced. Similar good results are obtained from the Parzival database when trained with this reduced number of lines and this new approach.