As the use of AI continues to grow, a consortium of researchers has created an institute designed to help establish partnerships between universities and businesses
Artificial intelligence (AI), which has exploded in the last decade, begins 2019 with even greater motivation in Brazil. The Advanced Institute for Artificial Intelligence (AI²), which will be officially inaugurated in February, will act as a bridge to encourage research partnerships between universities and businesses. The institute plans to promote research for a diverse range of applications, in line with the highly multidisciplinary nature of this field of computer science. “AI seeks to simulate human intelligence, based not only on computing, but also on biology, engineering, statistics, philosophy, physics, linguistics, mathematics, medicine, and psychology, just to name a few areas,” says computer scientist André Carlos Ponce de Leon Carvalho, deputy director of the Institute of Mathematical and Computer Sciences (ICMC) and head of the Data Analysis Machine Learning Research Center, both at the University of São Paulo (USP), and one of the members of the AI² institute.
The institute is represented by a consortium of artificial intelligence researchers from some of the largest universities in the country—all located in the state of São Paulo—which will offer their expertise to projects of academic and commercial interest. Physicist Sérgio Novaes, from the Scientific Computation Center at São Paulo State University (NCC-UNESP), is the coordinator of the group, and explains that the institute does not intend to limit itself to researchers from São Paulo or even Brazil: “Our goal is to establish partnerships around the world to foster research projects with high socioeconomic impacts.”
According to Novaes, the institute will be funded by its private partners. “The money will be used to recruit staff, organize events, subsidize travel by researchers, and eventually, to purchase software and hardware,” he explains. The aim is to establish collaborations that involve reciprocal and convergent interests so that AI² can stimulate important research, development, and innovation—and not just as mere service providers for the private sector. The institute will not have its own headquarters. “The model will rely on coworking spaces, sometimes at participating institutions, which will be interconnected via videoconferencing systems,” he says. Researchers, who will act as mentors to the private sector, will remain at their institutions, and company employees will continue to work from their offices.
It is not by chance that AI² was initially formed by researchers from São Paulo. A vast number of researchers, professionals, and businesses are based in the state, particularly in the city of São Paulo itself, which is a great appeal to technological development. The state is also historically home to a high number of researchers in this field. The USP Mathematics Department has seen AI research triple in the last three years, says computer scientist Roberto Marcondes Cesar Junior, a professor at the USP Institute of Mathematics and Statistics (IME) and head of the FAPESP Research, Innovation, and Dissemination Centers (RIDC) program. “Many of our students are recruited by startups in this sector,” he says. Ana Carolina Lorena, from the Computer Science Division of the Technological Institute of Aeronautics (ITA), São José dos Campos, agrees. “Undergraduate students and graduate researchers are increasingly being hired by companies to develop AI systems,” says the researcher.
Foreign researchers are attracted to Brazil by high research quality and good working conditions
Support from funding agencies such as FAPESP has been instrumental in how the sector has advanced in the state of São Paulo, especially through programs such as the Technological Innovation in Small Businesses (PIPE) program. “FAPESP supports advanced startups more than other funding agencies,” says Sérgio Queiroz, coordinator of research for innovation at FAPESP and a professor in the Institute of Geosciences at the University of Campinas (UNICAMP). Since 2012, the number of PIPE grants awarded to AI projects has grown significantly. This was confirmed by electronic engineer Marcelo Finger, head of the Computer Science Department at IME-USP. On November 26, at an event held by FAPESP and the São Paulo Legislative Institute—known as the ILP-FAPESP Cycle—Finger presented an overview of AI research in São Paulo. “FAPESP has been supporting artificial intelligence research since 1992. In 1997, when the PIPE program began, it was funding an average of five new projects a year in this field. In the last two years this number jumped to 40,” he says.
Brazil has even managed to attract a number of foreign researchers to the country due to its high research quality and good working conditions. One example is Englishman Brett Drury, who with PIPE funding is developing an intelligent risk analysis system for agriculture. “I came to Brazil because of the big agricultural market and the excellent reputation of USP and my advisor, Alneu de Andrade Lopes,” he says.
Drury says he met computer scientist Lopes at ICMC-USP in 2013, while doing his PhD at the University of Porto in Portugal, during which he developed a stock market forecasting system. During a postdoctoral fellowship funded by FAPESP, he swapped the stock market for commodities, studying sugarcane yields under Lopes’s supervision. Now, as head of research at SciCrop, based in São Paulo, he is developing a computational tool for agricultural risk assessment that analyzes text, satellite imagery, meteorology, historical production data, and even news articles on economic and social factors. “The system will help with hedging strategies [used to protect an investment against possible losses],” he says.
FAPESP is funding another event this year, organized by IME-USP and based on the São Paulo School of Advanced Science model, offering 10 mini-courses led by Brazilian and foreign researchers, each six to eight hours long. The courses will cover machine learning, statistics, databases, high-performance computing, and applications with a social impact.
Infographic Ana Paula Campos
Everyday technology Since the 1990s, AI projects supported by partnerships between FAPESP and businesses have resulted in many new products and industrial processes. One of the pioneers in the field is Claudio Oller do Nascimento, from the Chemical Engineering Department at USP. In 1997, a project developed in partnership with Petrobras through the Partnership for Technological Innovation (PITE) program resulted in an additional profit of US$0.25 per barrel of refined petroleum. “The refining process involves several variables, such as temperature and pressure. We used neural networks to adjust these parameters and optimize the process,” says the researcher.
A 2001 partnership with the São Paulo State Environmental Sanitation and Technology Company (CETESB) used neural networks to analyze meteorological data with the aim of predicting the formation of ozone in the lower atmosphere. “The ozone forecast was previously only available three hours in advance. The new system is capable of making predictions 24 hours beforehand,” he says. Oller is now one of the coordinators of the physical-chemical program at the Gas Innovation Research Center (RCGI), created in 2015 and funded by FAPESP and Shell.
Physicist Silvio Salinas, a former professor at the USP Physics Institute (IF-USP), now retired, was also among the first to receive FAPESP funding for AI research. “I never imagined that my research in the 1980s and 1990s would end up being linked to current AI activities,” he says. In the early 1990s, Salinas performed theoretical research in statistical physics that would later serve as the basis for the development of neural networks. “Our FAPESP projects contributed to the education of a good number of students. Nestor Caticha, for example, who is still at the Physics Institute, was another researcher in our group who analyzed neural network models, including decision-making and learning,” says Salinas.
Startup Treevia created a system to monitor forests remotely using sensors: the data is processed in real time by AILéo Ramos Chaves
After a period of relative obscurity, the study of neural networks has become one of the main lines of AI research since about 2010. “At the beginning of the century, scientists believed the area of neural networks was dead. But in the following 10 years, the game changed. Now, most AI applications are based on neural networks,” says Marcelo Finger.
Caticha, who is a member of AI2, says that the current explosion of AI research is due to the rising use of machine-learning algorithms, which have an increasing number of applications. This technology has already been incorporated into daily life to such an extent that it often goes completely unnoticed in everyday activities, reinforcing a quote attributed to one of the founding fathers of artificial intelligence, American computer scientist John McCarthy (1927–2011): “As soon as it works, no one calls it AI anymore.” Today, machine learning algorithms are used for smartphone facial recognition, biometric identification, and music and video recommendation algorithms.
I.Systems created a bottling system that uses AI to reduce wasteLéo Ramos Chaves
Accessible language According to computer scientist Ígor Braga, founder of Big Data, a São Paulo–based company that specializes in data science and machine learning, a growing number of businesses are looking to adopt AI solutions. One of the challenges is translating the results and decisions made by algorithms into accessible language. “Clients want to optimize their business, but they also want to know, as much as possible—why the algorithm makes the decisions it makes,” says Braga.
“Large companies tend to be more conservative about adopting new technologies, although that is changing,” says Esthevan Augusto Goes Gasparoto, CEO of Treevia, an agritech based in São José dos Campos in São Paulo State. With PIPE funding, the company created a web-based forest monitoring system that uses AI to process real-time data collected by sensors. “Every day, we see more companies with teams and initiatives dedicated to stimulating innovation [which seek knowledge outside their R&D departments]. That is where startups can contribute, by developing solutions at a speed that big corporations are not used to.”
Startups that develop innovations using smart tools are already beginning to feel the effects of high demand for professionals in the field. This is certainly the case for Stattus4, based in Sorocaba, São Paulo State. “We have had great difficulty finding qualified professionals to work on AI projects. And often when we do find them, their wage demands are high, which makes hiring difficult for small businesses and early-stage startups,” says Antônio Carlos Oliveira Júnior, the company’s technical director. This is a problem for companies looking for technology professionals. As a direct response, FAPESP created the TT4A and TT5 grant categories, which provide greater funding to those with proven computing experience.
Infographic Ana Paula Campos
Stattus4 created a system that detects water leaks using machine learning, a subarea of AI that focuses on the ability to recognize patterns and acquire knowledge through data analysis. The company received PIPE funding to help develop a prototype of its sensor, which records the vibrations caused by water flowing through pipes and identifies leaks by analyzing changes against an online database. “The PIPE program’s role in the development of this sector, by enabling startups to hire TT5 grant recipients, has been fundamental,” says Oliveira.
I.Systems, a company from Campinas that specializes in industrial process control and automation, was also backed by FAPESP in its early days. Founded by four former UNICAMP students in 2007, the startup first received funding from the foundation two years later, to develop industrial control software based on AI techniques. According to computer engineer Igor Bittencourt Santiago, one of the company’s founding partners (see special supplement “PIPE, 20 years of innovation”), the resulting software, called Leaf, was successfully adopted by the soft drinks industry, reducing waste during the bottling and canning stages. I.Systems has about 60 employees, mostly engineers, and more than 40 clients.
Cobli, another startup funded by PIPE, develops internet of things (IoT) solutions for fleet monitoring and management. The company has more than 70 employees, 20 of whom specialize in product development, and more than 600 customers across the country. It was founded just three years ago, when American entrepreneur Parker Treacy identified Brazil as a unique potential market and decided to invest in the logistics sector. He started the company with engineer Rodrigo Mourad using his own money and with no support from incubators.
Based in São Paulo, Cobli was awarded PIPE funding to improve a tool designed to identify driver behavior patterns via machine learning, using data obtained by solar-powered tracking devices. “With PIPE support we were able to establish a partnership with academia. AI has the potential to revolutionize the market through the innovative solutions it offers,” says technical director Lucas Fernandes Brunialti.
One of the areas that has invested most in AI is the health sector, where companies have developed various tools to support medical diagnoses. One example is Onkos Diagnósticos Moleculares, located in the Supernova Innovation and Technology Park in Ribeirão Preto, which has ties to USP. In 2018, Onkos received the 9th Octávio Frias de Oliveira Award for its diagnostic test that uses AI algorithms to classify metastatic tumors of unknown origin, called the Tumor Origin Test (TOT).
In addition to the TOT, the startup also uses AI in another diagnostic test to analyze indeterminate thyroid nodules. These nodules are traditionally investigated by collecting tissue through a procedure called FNA (Fine-Needle Aspiration). The results of the FNA procedure are inconclusive in anywhere from 15% to 30% of cases. The molecular test developed by Onkos is intended for patients who fall into this group. Called mir-THYpe, it uses AI to classify the indeterminate nodule as “negative” or “positive” for malignancy based on expression analysis of 11 microRNAs—small sequences of RNA that regulate gene expression. According to Onkos, the exam can reduce unnecessary thyroid removals by up to 81%.
Projects 1. Agricultural risk models based on textual information (nº 16/15524-3); Grant Mechanism Technological Innovation in Small Businesses program (PIPE); Principal Investigator Brett Mylo Drury (SciCrop Informação e Tecnologia S/A); Investment R$815.40. 2. Water leakage detection system using machine learning (nº 15/01100-4); Modalidade Technological Innovation in Small Businesses program (PIPE); Principal Investigator Antonio Carlos Oliveira Júnior (Stattus4); Investment R$77,623.40. 3. Predictive control of hydraulic pumps in water distribution systems through optimization of dynamic models (nº 15/08665-7); Grant Mechanism Technological Innovation in Small Businesses program (PIPE); Principal Investigator Ronaldo Antônio da Silva (I.Systems Automação Industrial); Investment R$535,991.58. 4. Advanced solution for identifying and evaluating drivers through large-scale machine learning and connected solar tracking (nº 16/08460-9); Grant Mechanism Technological Innovation in Small Businesses program (PIPE); Principal Investigator Lucas Fernandes Brunialti (Cobli); Investment R$479,196.01. 5. Integrated forest-monitoring system – Smartforest: technological revolution in forest inventory (nº 17/07593-8); Grant Mechanism Technological Innovation in Small Businesses program (PIPE); Principal Investigator Esthevan Augusto Goes Gasparoto (Treevia); Investment R$883,969.20. 6. Multicentric characterization and validation of a molecular examination for the classification of indeterminate thyroid nodules based on microRNA profiling (nº 17/16417-9); Grant Mechanism Technological Innovation in Small Businesses program (PIPE); Principal Investigator Marcos Tadeu dos Santos (Onkos Diagnósticos Moleculares); Investment R$731,330.86. São Paulo 7. School of Advanced Science on Learning from Data (nº 18/16488-6); Grant Mechanism Escola São Paulo de Ciência Avançada (ESPCA); Principal Investigator João Eduardo Ferreira (USP); Investment R$599,450.00
This article may be republished online under the CC-BY-NC-ND Creative Commons license. The Pesquisa FAPESP Digital Content Republishing Policy, specified here, must be followed. In summary, the text must not be edited and the author(s) and source (Pesquisa FAPESP) must be credited. Using the HTML button will ensure that these standards are followed. If reproducing only the text, please consult the Digital Republishing Policy.