Researchers from Universidade Estadual Paulista (Unesp), Bauru campus, in collaboration with colleagues from Friedrich-Alexander-Universität Erlangen Nürnberg, from Germany, intend, through a project supported by FAPESP, to optimize advanced artificial intelligence techniques used today so that , through algorithms, computers are able to collect data, interpret them and make predictions and generalizations from them.
The project was selected in the third call of 2016 of the São Paulo Researchers in International Collaboration (SPRINT) program, the result of which was announced at the end of January. It also counts with the participation of André Carlos Ponce de Leon Ferreira de Carvalho, professor at the Institute of Mathematics and Computer Sciences (ICMC) of the University of São Paulo (USP), São Carlos campus, and André Fujita, professor at the Institute of Mathematics and Statistics (IME) from the same university.
The call for the program, which aimed to promote the advancement of scientific research through collaborations in medium and long-term joint projects between researchers linked to universities and research institutions in the State of São Paulo and scientists from teaching and research institutions in the abroad, registered a record of selected proposals.
Twenty-four mobility proposals were selected between researchers from the State of São Paulo and from seven teaching and research institutions from five countries, with which FAPESP has a cooperation agreement, and four more proposals from researchers whose partners are linked to four institutions, from four different countries, with which the Foundation does not yet have an agreement in force.
The program has an open call until April 24 for the submission of new mobility proposals between researchers from the State of São Paulo and 14 teaching and research institutions from nine countries, with which FAPESP has a cooperation agreement.
“This is the third project that we have submitted and been selected within the scope of the SPRINT program,” João Paulo Papa, professor at the Department of Computing at Unesp in Bauru and the researcher responsible for the project on the Brazilian side, told Agência FAPESP.
“We had already carried out a project in collaboration with colleagues from the Ohio State University, in the United States, in 2015, and from the RMIT University, in Australia, in the same year, also in the area of optimizing advanced artificial intelligence techniques aimed, in the latter case, , for the imaging diagnosis of diabetic retinopathy [damage to blood vessels in retinal tissue caused by diabetes]. And now, we intend to use this same approach in bioinformatics [the application of information technology techniques aimed at analyzing and modeling data obtained in biological research]”, he said.
bioinspired algorithms
According to the researcher, one of the most advanced techniques used today in artificial intelligence to analyze large amounts of data and extract knowledge from them are those of the deep learning type.
Already used by companies such as Facebook to perform user recognition from photos, for example, one of the advantages of deep learning algorithms is their ability to learn large amounts of data in an unsupervised way (data not annotated, in computer jargon).
To do that, however, these algorithms have to deal with hundreds of parameters, Papa explained. “The challenge of creating deep learning algorithms is precisely choosing the most appropriate parameters, because each application requires a different configuration”, he said.
One of the approaches that have been adopted to identify the best desired parameters for a given application is to run an algorithm numerous times to select the best result.
In order to reduce the time of this process of random choice of parameters, the researchers intend to evaluate the use of algorithms called bioinspired, so called because they are inspired by nature. One of them, for example, is based on the behavior of ants.
Ants tend to choose the shortest path during their travels because, as they walk, they release pheromones so that other members of their colony can follow them. And if the route they choose is long, there is a risk that the pheromone they released will dissipate and the ants in their colonies will lose these tracks, Papa explained.
“Algorithms that are bioinspired by the behavior of ants are based on this premise to choose the best parameter values for a specific application in a feasible time, in order to optimize or reduce the error of an application of an artificial intelligence technique”, he detailed.
The researchers chose the area of bioinformatics to validate the use of these techniques because collaborators from USP and the German university were interested in this area.
One of the applications they envision is for comparing the DNA of two people, for example, to analyze their levels of similarity.
“The aim of our project is to show in a concrete example in the field of bioinformatics that machine learning and optimization methods can be used in synergy,” said Alexander Martin, professor in the Department of Mathematics at Friedrich-Alexander-Universität Erlangen Nürnberg and researcher responsible for the project abroad.
In his opinion, FAPESP's SPRINT program will make it possible to finance the exploratory phase of the project, which is often difficult to carry out through programs maintained by other research funding agencies in Brazil and Europe, he compared.
“The SPRINT program will allow us to continue in this line of research with the great collaboration of our colleagues in São Paulo”, he said.
More information about the SPRINT Program can be found at www.fapesp.br/sprint.
EN 
PT 
