Short Bio

I joined UNSW as an associate professor in 2019 after working for more than ten years at the University of Sao Paulo (USP). During 2010-2012, I was a visiting researcher at the University of California, Riverside (UCR), working in the prof. Eamonn Keogh's laboratory.

During my stay at UCR, I continued my work with time series analysis, particularly developing methods for classifying and clustering time-oriented data. In conjunction with Dr Keogh, I proposed the first time series distance invariant to complexity and speed-up techniques to compare massive amounts of time series data under warping.

More recently, I have worked with data streams, particularly with classification with label latency and proposed efficient unsupervised methods to detect concept drifts as well as to learn in the presence of these changes in the data distribution.

My research is motivated by applying Machine Learning in practice. My approach is to work on challenging applications that help my students and me to identify gaps in the literature or assumptions in the state-of-the-art that do not hold for our applications. This research approach often leads to contributions in Computer Science and the application areas.

One such approach is the challenge of incorporating classification algorithms on embedded devices. For example, I have developed lightweight models that can run in environments with severe power restrictions, such as satellites and sensors. One notorious application is the development of sensors to classify insects in flight automatically, allowing the creation of surveillance systems for disease vectors, invasive species and pests. I have also developed EmbML, a Machine Learning tool to convert sickit-learn and Weka classifiers into C++ code crafted to run into low-power microcontrollers, such as ones found in the Arduino family.

In the last years, I have actively worked in Machine Learning Quantification, developing new algorithms to count events accurately. These recent developments have led to the proposal of a novel Data Mining task known as One-class Quantification and a family of efficient quantification algorithms.