Hector Javier Hortua

Biography

The research I am currently involved in consists of the application of Deep learning in several scientific disciplines and in real world-dataset, along with the study of probabilistic inference through Markov Chain Monte Carlo (MCMC) and Variational Inference (VI). More specifically, my research falls under two topics: First, the use of neural ensembles, Bayesian neural networks and Gaussian Processes for modeling uncertainties in Deep learning, which are crucial for making better decisions in real-world applications. Second, the implementation of alternative generalized divergences and Bijectors in VI in order to improve the inference processes, and be able to obtain well-calibrated neural networks. I have also worked on the robustness of Bayesian classifiers for detecting adversarial examples through Variational AutoEncoders or calibrated networks, and the implementation of stochastic neural networks for Object Detection in computer vision methods. Finally, I have built different end-to-end data science/ML projects starting from building an ML-focused strategy and progressing into model training, optimization, and productionalization using Google Cloud Platform.