Applications in AI developed in PB are used nationally

publicado: 24/01/2020 12h00, última modificação: 03/03/2020 09h16
All the techniques mentioned will be taught in the new course Data Science and Artificial Intelligence at UFPB
1 | 2
Teacher Yuri. Photograph Diego Nóbrega.
2 | 2
Teacher Gilberto. Photograph Diego Nóbrega
Photograph Diego Nóbrega
Photograph Diego Nóbrega

It's not science fiction. It's scientific reality. Artificial Intelligence is a computational resource built and dominated by the human being. It brings together a set of techniques that can be applied to solve various problems. The reporting team visited the Computer Center of the Federal University of Paraíba and visited the laboratories where these solutions are created and developed. From their ownership, citizens, companies, governments, experience the benefits. All the techniques mentioned will be taught in the new course Data Science and Artificial Intelligence at UFPB.


Portuguese to “Libras” translator - VLibras


The application that helps deaf people translate texts from Portuguese into the Brazilian Sign Language has learned to differentiate the meaning of words with more than one meaning. VLibras is a free resource for computers and mobile phones that reads texts on the Web and, through a nice little doll, the avatar, transforms the content in gestures understandable to the deaf. "Most of them have difficulty reading Portuguese because ‘Libras’ is their first language; Portuguese is like a foreign language," explains Tiago Maritan, of the Digital Video Lab (Lavid), which since 2009 has been developing with students the application created at the UFPB's Computer Centre.


"We always used artificial intelligence to develop VLibras, but it was a classic approach, not based on machine learning. We are dealing with the limitation of the aspects related to the context in which the words are inserted. Is orange the fruit, is it the color, or is it a person who behaves like an 'orange'? In 2019 we assembled a team of interpreters to build a database of over 100,000 sentences - translation examples - and applied machine learning to VLibras. In December we released the updated version of VLibras," explains the researcher. VLibras is used in government sites, such as the website of the Government of Paraíba, and is available for free on the Internet.


Credit Analysis


The time to get the answer to a credit request in a store will be almost instantaneous with the entrance of artificial intelligence in that process.


Artificial intelligence learns from the history of credit applications that have been granted or denied in the store and identifies the profile of good payers. What's more, of those who have had credit, the machine will know the profile of who paid, who delayed, or is in debt.


When a new client applies for credit, it will be possible to tell quickly if he will have approval or not.


Gilberto Farias, Head of the Scientific Education Department of the Computer Science Center of the UFPB and the Laboratory of Systems Engineering and Robotics (Laser), informs that the tool used is the "machine learning in the application of credit". This application is under development and will soon start to be used in a retail store in Paraíba. The app will be automated, the risk of default will be lower and customers will gain time.


Route optimization for product deliveries


Deliveries to retail stores will be made more efficiently on pre-programmed routes. Artificial intelligence will calculate the best sequence for deliveries to various addresses.


The technique used was that of algorithms with metaheuristics, specific for optimization problems. It is the same as the "distance between two points", but to solve the routing problem it is more difficult: there is a sequence of distances that need to be combined to give the shortest route. Google Maps gives routes between two points within a sequence that the user enters. The system developed at UFPB goes further, it determines the sequence of addresses by making the shortest route, within a scheduled time window.


"With this system we were able to lower by 30% the route that all the trucks took in one day of testing. The fleet of 6 trucks has been reduced to 5, without exceeding the drivers' schedule," informs Gilberto Farias (Laser). The application is already in use in a retail store in Paraíba.


Project of IA speeds up processes in the Court of Auditors of the State of Paraiba


One of the tasks of the TCE auditors is to analyze the commitments sent by the municipalities to assess whether the expenditure is in agreement. The commitment is a promise of payment, but the TCE only releases if the commitment refers to the money allocated - a payment of student uniforms with the education money, for example.


The TCE receives a large volume to analyze and they do it manually, one by one. The response for the municipality takes time and harms citizens. This analysis can be done faster and less errors by machines and speed up public administration. Yuri Malheiros, from ARIA, Artificial Intelligence Applications Laboratory (CI/UFPB), explains that the machine is learning the patterns with analyses of an old process data collection. "We use the recurrent neural network technique," she explains.


"To get the machine to learn an AI technique, it is necessary to consult previous data. And to extract the information, apply machine learning. It's Data Science associated with Artificial Intelligence," speaks Malheiros.


Intelligent inventory configuration


When a store launches an advertising campaign with unmissable promotions the consumer is motivated to buy. What if, upon arrival at the store, the product has run out? This is the so-called "out of stock" - successful advertising finds no support in the product's stock planning and sales fail.


Artificial intelligence corrects that flaw. In this case, the program makes a projection of what could have been sold if there was still the stock. A simple mathematical technique is applied. The system looks at the sales curve. Suddenly it zeroes out - the rupture occurs - the product is over. Artificial Intelligence makes an interpolation in this straight, as if it continues following until the date that the stock is put back in the store. In that vacuum, how many products would be sold if we didn't run out of stock? This is the answer to the quantity of stock to be planned.


Through the "random decision tree" technique, the machine learns to generate corrected

stock based on past history.





Translated by Iohan Faustino