10/12/2019

Variety of topics marks the Applied Data Science course.

Vitor Guerra durante apresentação

What could a round of the Brazilian Championship, deforestation, and viral emails possibly have in common? The odds of all these phenomena occurring can be decoded and meticulously analyzed through Applied Data Science. This was investigated during a course taught for the first time at IMPA by postdoctoral researcher Vitor Guerra, from Visgraf (IMPA's Computer Graphics Laboratory).

“It’s a very current discipline that reinforces IMPA’s pioneering spirit. The content began to be planned about two years ago, and the course transcends the course itself. Today, there is a high demand in the market for professionals who know how to apply the techniques we are teaching here,” says Luiz Velho, the lead researcher of Visgraf.

Through various methodologies related to the field of Data Science, the students analyzed results from the ENEM (National High School Exam), musical frequency, CNPJ data, the origin of fake news , and deep learning . Some students arrived at results with social applicability. For example, student Kizzy Terra found direct correlations between the educational levels of mothers and children: the higher the mother's level of education, the greater the student's chance of passing the ENEM exam.

The classes were held between August and November for master's and doctoral students. The class had sixteen participants, who presented their work on November 2Thursdays and 28th to conclude the semester. All videos of the final work presentations are available on the course website .

Apresentações finais dos alunos estão disponíveis

“The participation of monitor Pedro Arthur was essential to increase student engagement. We worked with very current topics that he brought to share with the students, which was very positive. It increased engagement in the classes,” says Vitor.

Carolina Piazza opened the first day of presentations with two papers. The IMPA master's student spoke about possible variant causes for the Amazonian deforestation recorded between 2014 and 2016. Following that, she analyzed natural language processing (NLP) to evaluate word repetition in scientific publications. Rafael Westenberger presented a paper on the exchange of corporate emails and the dissemination of words, attempting to predict the chances of an email going viral within a group.

On the second day of presentations, nine doctoral students presented their final projects for the course. Thiago Ramos, for example, spoke about variational autoencoders (VAEs), and Julia Lemos about dynamical systems. Another topic presented was the use of generative adversarial networks ( machine learning ).