Academia and private enterprise join forces to foster research in artificial intelligence12 de março de 2019
By Elton Alisson | FAPESP Research for Innovation – Centers have been established in recent years in major cities such as New York and Toronto to conduct research in artificial intelligence (AI) by means of collaborations between academia and private enterprise. A new AI research center has just been established in São Paulo, the largest city in the southern hemisphere.
A group of researchers affiliated with eight universities in São Paulo State, in partnership with major corporations and startups in various industries, launched the Advanced Institute for Artificial Intelligence (AI²) on February 26, 2019.
The main aim of the institute is to foster interactions among researchers dedicated to developing advanced problem-solving computational methods and devices and exploring applications of these that can have significant social and economic impacts.
The institute plans to use coworking, which is a network of shared workspaces connected by a videoconferencing system in participating organizations.
“The academic community will contribute experience in basic research on AI, while the private sector will offer challenging problems and financial support for us to work together on projects with potential socioeconomic benefits,” said Sérgio Novaes, Full Professor of Physics at São Paulo State University (UNESP) and one of the founders of AI², during the ceremony held to launch the institute.
Initially, the consortium will comprise researchers from the University of São Paulo (USP), the University of Campinas (UNICAMP), São Paulo State University (UNESP), the Federal University of São Paulo (UNIFESP), the Federal University of the ABC (UFABC), Mackenzie Presbyterian University, FEI University Center, and the College of Engineering and Management (ESEG).
The private-sector partners include IBM, Intel, Petrobras, Grupo Fleury and Serasa Experian. The institute is interested in partnering with other business organizations, universities and research institutions in Brazil and abroad.
During the launch event, a workshop on opportunities for collaboration was held at UNESP’s Scientific Computation Center in São Paulo.
“We want to facilitate interaction between academia and business for the development of AI projects,” Novaes said.
One of the first companies to join the initiative, Serasa Experian, announced that it will award three postdoctoral fellowships for research on alternative data, chatbot technologies and AI-driven credit analysis.
“We’re interested in using machine learning technologies to acquire data relevant to shopping decisions, ranging from how consumers choose a product to how they finance their purchases and what they do to pay off their debts,” said Renato Vicente, a professor at the University of São Paulo’s Mathematics and Statistics Institute (IME-USP) and data science director of Serasa Experian’s Data Lab.
Scientists affiliated with AI² are currently working on three lines of research funded by private-sector partners. One of the projects aims to improve the monitoring of oil wells and develop computer vision software to identify hazardous situations involving offshore platforms.
Another aim is to promote the use of robotics as an educational tool, develop service robots that do useful work for humans or equipment, excluding industrial automation applications, and study human-machine interaction.
The third project aims to accelerate simulations of atomic particle collisions using machine learning techniques with a range of applications, such as distinguishing between a true photograph and a montage.
Simulations are essential to the verification of real data from particle collisions in the Large Hadron Collider (LHC) at CERN, the European Organization for Nuclear Research. To obtain the required statistical accuracy, two-to-three minute collision simulations must be run. The simulation challenge will become even more complex when the LHC upgrade goes live this year or next year. Between 30 and 40 concurrent collisions can be recorded right now, but after the upgrade, the number will be in the range of 200.
“This will invalidate the way we currently do simulations. We’ll need computers that can run the data six times faster, or we’ll have to multiply the number of computers we use for this purpose by six,” physicist Thiago Tomei told Agência FAPESP.
Thomei is affiliated with SPRACE, the São Paulo Research and Analysis Center, which was founded in 2003 with FAPESP’s support. Coordinated by Novaes, SPRACE enables researchers in high-energy physics from São Paulo State to participate in the LHC’s Compact Muon Solenoid (CMS) experiment.
The researchers at AI² plan to use machine learning techniques to accelerate simulations a thousandfold, speeding up the identification of particles after a collision event and increasing the precision of the results.
“The idea is to use generative models to enable algorithms to ‘learn’ to classify data samples and, by doing so, generate new samples,” Tomei explained.
Area in ascent
AI was long associated with solutions capable of making machines and software learn to perform tasks based on human reasoning. However, experts in the field explain that an AI revolution has been driven by several factors in recent years, including big data and cloud computing, which have both enabled scientists to develop algorithms for specific tasks, such as driving an autonomous vehicle, identifying cancer cells in MRI scans or detecting crop pests.
China, Japan, France, Canada, the US and the UK, among other countries, have invested heavily in AI because they see it as an area of growth that is strategic to economic, social and scientific development and is required to guarantee homeland security, independence and national sovereignty.
China, for example, plans to invest US$1 trillion in AI by 2030, aiming to ensure its leadership of the field. Canada has two AI innovation research centers, in Toronto and Montreal, and in 2017, it launched a national policy to stimulate AI development.
“Canadian investment in AI has increased strongly in recent years. The number of AI startups in Canada rose 30% in 2018,” said Marie-Hélène Béland, Canadian Vice Consul in São Paulo, during the event.
Brazilian investment in AI has been timid according to participants in the event who said the establishment of AI² will boost the development of this knowledge area.
“Unfortunately, Brazil is lagging behind in this area. If we fail to keep up with progress in AI, we risk being left behind for good. It will be worse than in the case of semiconductors, where Brazil missed the boat and has never been able to catch up,” Novaes said.
“One way to stay in the AI game is to make sure we’re continuously up to date and at the same time start building direct links between the private sector and academia to develop joint projects.”
The researchers affiliated with the institute believe that the support of FAPESP and other research funding agencies has been decisive for the development of this area in São Paulo State, especially via FAPESP’s Innovative Research in Small Business Program (PIPE).
In the past six years, the number of AI startups supported by PIPE has increased more than sixfold. In 1997, when PIPE was launched, the number of new AI projects supported annually averaged five. In the past two years, the number jumped to 40, according to data presented by Marcelo Finger, Full Professor of Computer Science at IME-USP and a member of FAPESP’s Computer Science Area Panel, during the ILP-FAPESP Cycle of Lectures on Artificial Intelligence held in November 2018 at the São Paulo State Legislative Assembly (ALESP).
According to Finger, FAPESP’s support for AI projects began in 1992 and has increased steeply since 2012.
In partnership with IBM, FAPESP supports the creation of an AI center under the aegis of the Foundation’s program of Engineering Research Centers (ERCs). A call for proposals has been issued, targeting research groups affiliated with universities and research institutions in São Paulo State.
On March 12, 2019, FAPESP and IBM held a meeting at the Foundation’s headquarters to explain the call, with presentations on AI development in several sectors. The closing date for the submission of proposals is April 15, 2019.
Specific instructions for researchers in São Paulo State can be found at: www.fapesp.br/en/12508.