Jorge Vergara

July 26th – October 26th, 2017.

Jorge is a machine learning expert who works on feature selection, identifying which features provide the most information in a data set.

Jorge was born in Temuco, Chile. He obtained his B.S. and P.E. degrees in Electrical Engineering from the University de La Frontera in 2005 and 2007 respectively.

Jorge obtained his Ph.D. in electrical engineering from the University of Chile in 2015 where he worked on feature extraction and selection method based on information theory. Actually Jorge is a postdoctoral research at the University of Chile and the Millennium Institute of Astrophysics, where he works on selection and extraction of feature group based on mutual information for classification of patterns in astronomical images and time series.

Jorge’s scientific interests are interaction and causality in features selection method using information theoretic, quantization of nonlinear time series and hierarchical learning. The Jorge’s current work focuses in detection of candidate asteroid on stamp images in Moving Object Pipeline System on LSST and dimensionality reduction and feature-learning in astronomical spectroscopic data using Variational Autoencoder.

Working at the UW

In Jorge’s stay at the University of Washington, he works with Andrew Connolly’s team on two main topics:

1)  Study of new strategies to optimize the MOPS process in real time to detect asteroid. In this study they work with unsupervised analysis of images (stamp) and hierarchical classification where they create and select the most relevant individual and group features to discriminate between asteroids and non-asteroid.

2)  Feature Extraction from spectroscopic data using Variational Autoencoder. In this study they work on the dimensionality reduction in spectroscopic data using Variational autoencoder to reconstruct spectroscopic data with with different resolution and to map nonlinear similarities on spectra.