César Roberto de Souza

Procedural Generation of Videos to Train Deep Action Recognition Networks

Deep learning for human action recognition in videos is making significant progress, but is slowed down by its dependency on expensive manual labeling of large video collections. In this work, we investigate the generation of synthetic training data for action recognition, as it has recently shown promising results for a variety of other computer vision tasks. We propose an interpretable parametric generative model of human action videos that relies on procedural generation and other computer graphics techniques of modern game engines. We generate a diverse, realistic, and physically plausible dataset of human action videos, called PHAV for Procedural Human Action Videos. It contains a total of 39,982 videos, with more than 1,000 examples for each action of 35 categories. Our approach is not limited to existing motion capture sequences, and we procedurally define 14 synthetic actions. We introduce a deep multi-task representation learning architecture to mix synthetic and real videos, even if the action categories differ. Our experiments on the UCF101 and HMDB51 benchmarks suggest that combining our large set of synthetic videos with small real-world datasets can boost recognition performance, significantly outperforming fine-tuning state-of-the-art unsupervised generative models of videos.

A neural network approach to predicting outcomes in heart failure using cardiopulmonary exercise testing

OBJECTIVES To determine the utility of an artificial neural network (ANN) in predicting cardiovascular (CV) death in patients with heart failure (HF). BACKGROUND ANNs use weighted inputs in multiple layers of mathematical connections in order to predict outcomes from multiple risk markers. This approach has not been applied in the context of cardiopulmonary exercise testing (CPX) to predict risk in patients with HF. METHODS 2635 patients with HF underwent CPX and were followed for a mean of 29 ± 30 months. The sample was divided randomly into ANN training and testing sets to predict CV mortality. Peak VO2, VE/VCO2 slope, heart rate recovery, oxygen uptake efficiency slope, and end-tidal CO2 pressure were included in the model. The predictive accuracy of the ANN was compared to logistic regression (LR) and a Cox proportional hazards (PH) score. A multi-layer feed-forward ANN was used and was tested with a single hidden layer containing a varying number of hidden neurons. RESULTS There were 291 CV deaths during the follow-up. An abnormal VE/VCO2 slope was the strongest predictor of CV mortality using conventional PH analysis (hazard ratio 3.04; 95% CI 2.2-4.2, p<0.001). After training, the ANN was more accurate in predicting CV mortality compared to LR and PH; ROC areas for the ANN, LR, and PH models were 0.72, 0.70, and 0.69, respectively. Age and BMI-adjusted odds ratios were 4.2, 2.6, and 2.9, for ANN, LR, and PH, respectively. CONCLUSION An ANN model slightly improves upon conventional methods for estimating CV mortality risk using established CPX responses.

Fingerspelling Recognition with Support Vector Machines and Hidden Conditional Random Fields

In this paper, we describe our experiments with Hidden Conditional Random Fields and Support Vector Machines in the problem of fingerspelling recognition of the Brazilian Sign Language (LIBRAS). We also provide a comparison against more common approaches based on Artificial Neural Networks and Hidden Markov Models, reporting statistically significant results in k-fold cross-validation. We also explore specific behaviors of the Gaussian kernel affecting performance and sparseness. To perform multi-class classification with SVMs, we use large-margin Directed Acyclic Graphs, achieving faster evaluation rates. Both ANNs and HCRFs have been trained using the Resilient Backpropagation algorithm. In this work, we validate our results using Cohen’s Kappa tests for contingency tables.

Predicting reintubation, prolonged mechanical ventilation and death in post-coronary artery bypass graft surgery: a comparison between artificial neural networks and logistic regression models.

Introduction In coronary artery bypass (CABG) surgery, the common complications are the need for reintubation, prolonged mechanical ventilation (PMV) and death. Thus, a reliable model for the prognostic evaluation of those particular outcomes is a worthwhile pursuit. The existence of such a system would lead to better resource planning, cost reductions and an increased ability to guide preventive strategies. The aim of this study was to compare different methods – logistic regression (LR) and artificial neural networks (ANNs) – in accomplishing this goal. Material and methods Subjects undergoing CABG (n = 1315) were divided into training (n = 1053) and validation (n = 262) groups. The set of independent variables consisted of age, gender, weight, height, body mass index, diabetes, creatinine level, cardiopulmonary bypass, presence of preserved ventricular function, moderate and severe ventricular dysfunction and total number of grafts. The PMV was also an input for the prediction of death. The ability of ANN to discriminate outcomes was assessed using receiver-operating characteristic (ROC) analysis and the results were compared using a multivariate LR. Results The ROC curve areas for LR and ANN models, respectively, were: for reintubation 0.62 (CI: 0.50–0.75) and 0.65 (CI: 0.53–0.77); for PMV 0.67 (CI: 0.57–0.78) and 0.72 (CI: 0.64–0.81); and for death 0.86 (CI: 0.79–0.93) and 0.85 (CI: 0.80–0.91). No differences were observed between models. Conclusions The ANN has similar discriminating power in predicting reintubation, PMV and death outcomes. Thus, both models may be applicable as a predictor for these outcomes in subjects undergoing CABG.

Artificial Neural Networks Prognostic Evaluation of Post-Surgery Complications in Patients Underwent to Coronary Artery Bypass Graft Surgery

In this paper we explore the applications of artificial neural networks in the field of heart surgery, more specifically in the prognostic evaluation of post-surgery complications, such as death, reintubation, prolonged mechanical ventilation and the need for extracorporeal circulation in patients who underwent coronary artery bypass graft surgery. Predictive variables were limited to information available before the procedure, and outcome variables were represented only by events that occurred postoperatively. We also employed the principal component analysis technique to further reduce the complexity of our input data set in an attempt to improve artificial neural network efficiency and reliability