I currently work as an Associate Professor at the Institute of Mathematics and Computer Sciences, Department of Computer Science, University of São Paulo, São Carlos, Brazil. I obtained my PhD degree in Electrical Engineering from the University of São Paulo, São Carlos in november 2003. I worked with Real-Time Operating Systems during my Masters, and with High-Performance Computing (HPC) in my PhD.
From March 2004 to 2005, I decided to start studying Machine Learning, Time Series Analysis, and Applied Dynamical Systems still motivated by applications in the area of HPC. That was the best experience ever! As soon as I covered most of the basic concepts I decided to go for applications and the theoretical foundation. That was the way I migrated to the area of Machine Learning. I still study and apply Dynamical Systems and Time Series Analysis in my work too. In summary, my research interests include Machine Learning, Applied Dynamical Systems, Time Series Analysis, and Data Streams.
I love researching and teaching!
I have been writting a book on the Statistical Learning Theory with the cooperation of Prof. Moacir Ponti (also working in the same Department at the University of São Paulo). This book will bring practical examples and codes written for the R Statistical Software.
With the support of CNPq (National Council for Scientific and Technological Development) we have been working since july 2014 to provide a complete platform to listen to Twitter streams and process them out using Dynamical Systems, Time Series and Machine Learning tools. The main idea behind this project is to mine data streams on-the-fly and provide a Web tool for end users to know what is happening to certain Twitter hashtags. In our case, we are collecting and processing hashtags associated to politics, health and other social interests.
Phase-space embedding aims at finding the best immersion for some time series or data stream. The main idea behind this project supported by FAPESP (The São Paulo Research Foundation) is to unfold time-dependent data into the best as possible space which should help modeling and predicting such type of data. There are several applications that could take advantage of this project, such as the study of population growth, climate studies, stock market modeling/prediction, etc.
We have been working on how to provide theoretical support for the selection of the number of neurons per layer of Convolutional Neural Networks (CNNs), as well as their number of layers and sizes of the convolutional kernels. As main objective, we wish to contribute on building up simpler and more adequate CNN architectures to tackle real-world problems.
We have been working hard on how to provide theoretical guarantees for the detection of concept drifts in data streams. We have already some relevant published results, but we are still working on new stuff. The basic idea behind this project is to point out when significant changes occur on a continuously collected stream of data, having guarantees though.
As we have the Statistical Learning Theory to provide theoretical foundation for the supervised learning, we have been working on how to provide stronger theoretical guarantees for unsupervised learning.