Give solutions to problems and provide an explanation of phenomena that have occurred, but also anticipate those that may happen. This is what technology and science are all about and summarizes, in turn, the aims that the
two research projects of Argentine scientists which, this week, resulted
winners of the Latin America Research Awards (LARA), delivered to the Engineering Center for Latin America of Google, in the city of Belo Horizonte, in Brazil.
Francisco Soulignac and the doctoral student
Gonzalo Lera-Romero, both from the Computer Department of the University of Buenos Aires on the one hand; and
Ana Gabriela Maguitman and the doctoral student
Mariano Maisonnave, from the Institute of Computer Science and Engineering of the National University of the South, on the other; They are the ones who received a scholarship to finance their research for 12 months. Another 23 projects (15 from Brazil, 5 from Colombia, 2 from Chile and 1 from Per) were also winners of the LARA, out of 670 applications. In this edition, 500,000 dollars were allocated to support the scientific works.
How to improve merchandise deliveries
Francisco Soulignac and Gonzalo Lera-Romero investigate how to improve the planning of direct deliveries of merchandise to consumers in online purchases Credit: Shutterstock
The project presented by the UBA Computing Department researcher Francisco Soulignac and Gonzalo Lera-Romero (in which Juan Jos Miranda Brown, from the Di Tella University also participates) addresses a problem that is of special interest in custom logistics that the electronic commerce grows: how to improve the planning of the direct deliveries of merchandise to the consumer in the so-called "last mile" – the final stage of the chain of distribution of the products – taking into account the congestion that could be in the transit.
Improving the current algorithms to incorporate the congestion factor, and thus facilitate decision-making on the order of product distribution, is the main challenge. "Today there are applications that we all use: we say 'I am here, I want to go here', and they give us a route. That is reasonably well solved, it is simple in computational terms, the problem is when one has to visit many locations" said Soulignac.
But if a company has to do, for example, one hundred deals in different parts of a city, the problem "explodes from combinations" and there are many difficult possibilities to analyze one by one, explains the researcher.
The problem, known as "vehicle routing", becomes more difficult to solve if factors such as the times when a customer can be visited are taken into account or if a vehicle or more than one is owned for delivery. And even more complex if this traffic is incorporated into this calculation.
"The e-commerce so far in Argentina did not explode at all, but the need to solve this type of problem begins to arise," said Lera-Romero. "The final objective that we hope to get out of this project is to have a tool that helps companies make decisions," he added.
Having a tool that provides an adequate order to visit customers taking into account congestion will not only favor companies in terms of costs but also help not add more traffic jams to the streets and not increase or reduce the carbon footprint that generates the logistics ..
While the project for now is limited to the academic field and has not been tested outside of this, a next step is to "polish" the algorithms to test them with data taken from reality.
Explain and predict the economy
Ana Gabriela Maguitman and Mariano Maisonnave, from the National University of the South, were distinguished by a research project on causal models that use data extracted from the media Credit: Shutterstock
Meanwhile, the researcher Ana Gabriela Maguitman and the PhD student Mariano Maisonnave, from the National University of the South, were also distinguished by a research project on causal models that use data extracted from the media.
The idea came from the request of a group of professors from the department of economics of the university. "There is much interest on the part of economists and financiers of trying to understand a complex event in terms of what variables are involved, and how they are linked to each other," said Maisonnave. "They wanted to work on these issues. They know tools of econometrics and are specialists in statistical matters, and we have the tools from the computational point of view. Then this idea of ??collaborating arose," said the specialist.
The goal of the research, Maisonnave said, is "to find a causal or interconnection graph to try to explain whether a certain connection between the variables impacts that this happened in one way or another." And he added: "The idea is that with a look at this graph an expert can have a more general idea of ??what happened, or see in the graph things that he did not realize because he would allow it from large volumes of data make a general summary ".
"That graph that we want is going to have nodes, which in addition to events we want to be variable, that come from other sources, not only from digital media," said Maguitman. Events such as bankruptcy, debt taking, the resignation of a president or the abrupt fall or rise of an index are some of the elements extracted from a corpus of the newspaper
The New York Times, with those who seek to train a neural network to first recognize events on their own and then know how to identify them to make a causal model. Although Maisonnave explained that the project is currently aimed at explaining events, the model may also be ahead of others. "In principle, what we want to do is build this graph as a way of explaining, but we also have the idea of ??moving forward on the path to predict."
In addition, in the future the model may add the moods to the graph. "One of the original ideas was eventually to incorporate them because one can do an analysis on a social network and detect if on a certain subject people are reacting with anguish, with joy, with fear." The model also implies the incorporation of events that do not have an economic origin, such as a natural disaster. "Many times a non-economic event, such as an earthquake, triggers an economic event," said Maguitman.
Of the 25 winning projects of the LARA, half of these point to the field of medicine. "There is a trend towards the application of technology in the area of ??health," said Berthier Ribeiro-Neto, director of Google engineering for Latin America, presenting three research projects that use machine learning and artificial intelligence (AI) to The medical diagnosis. Ribeiro-Neto stressed that, beyond the use of "deep learning" and AI, "people have to make decisions."
One of the projects, presented by Sandra Avila and Alceu Bissotto (State University of Campinas, Brazil), aims to improve the classification of skin cancer with an extension to the Antagnica Generation Network (GAN), a form of artificial intelligence. The scientists introduced a method based on GAN to generate realistic synthetic data with the purpose of improving injury classification models. Avila explained that the technology does not seek to make a diagnosis but "provide support."
On the other hand, Winston Percybrooks and Pedro Narvaez (Universidad del Norte, Colombia) were selected for the project "Towards a large-scale intelligent computer-assisted auscultation for remote primary care settings". Cardiac auscultation will be performed through a digital stethoscope operated on a cell phone. The sounds will be sent to the cloud so that a professional can evaluate them. The model is in the testing phase. "We expect the impact to be a lot in remote areas," Percybrooks said.
Meanwhile, researchers Mirko Zimic and Macarena Vittet (Universidad Peruana Cayetano Heredia) carry out an autism diagnosis project in children by combining preference in the eyes, pupilometry and recognition of emotional gestures, through mobile devices. The idea is based on the artificial vision of computers and deep learning algorithms to perform three analyzes based on the video of the child's face, recorded with the front camera of a tablet. "The methods we can use in machine learning and telemedicine are opportunities for screening," said Zimic, who added that he should then move on to a confirmatory stage.
. Google rewarded the research of four Argentine scientists in logistics and economics – LA NACION