Dos proyectos colombianos, ganadores en los ‘Premios de Investigación de Google para América Latina’

agosto 26 ,2017 0
Compartir esta noticia

Investigadores de la Universidad de los Andes fueron reconocidos por su proyecto  de dispositivos capaces de leer las emociones a través del reconocimiento de las micro expresiones faciales. El proyecto de telemedicina de investigadores de la Universidad del Norte, que busca construir a través de teléfonos móviles un estetoscopio para diagnosticar condiciones cardiacas en comunidades remotas, también fue uno de los ganadores.

Por quinto año consecutivo, Google anunció los ganadores de los Premios de Investigación de Google para América Latina (LARA, por sus siglas en inglés), entre los que fueron reconocidos dos proyectos colombianos.

El primero de ellos es de los investigadores de la Universidad de los Andes, Pablo Arbeláez y Andrés Romero, que recibieron por tercer año consecutivo un incentivo económico para continuar profundizando su investigación que busca tener dispositivos capaces de entender las emociones de las personas a través del reconocimiento de las microexpresiones faciales.

Por otra parte, por primer vez, dos investigadores de la Universidad del Norte de Barranquilla fueron reconocidos en estos premios; se trata de Winston S. Percybrooks y Pedro Juan Narváez Rosado cuyo proyecto de investigación en telemedicina busca construir a través de teléfonos móviles un estetoscopio de bajo costo, capaz de grabar y  clasificar latidos de corazón y con el uso de aprendizaje automático (machine learning) brindar diagnósticos de condiciones cardiacas en los pacientes, y apoyar atención médica no especializada, especialmente en zonas remotas.

En la ceremonia de premiación, además, se anunciaron otros 25 proyectos de investigación en Ciencias de la Computación que recibirán, durante los próximos dos años, 600 mil dólares para llevar a cabo sus iniciativas. Desde que fueron lanzados los Premios Latin American Research Awards (LARA) en el año 2013, el programa ha beneficiado a 46 proyectos y a más de 100 investigadores, incluyendo estudiantes de maestría y doctorado y sus profesores.

Este año se recibieron 281 proyectos de nueve países de América Latina. Es significativo y muy positivo el crecimiento en el número de investigadores de toda la región enfocados al estudio del aprendizaje automático, un campo que permite enseñar a las computadoras a aprender sin que éstas sean explícitamente programadas. De los 27 proyectos elegidos, 14 se enfocan en aprendizaje automático o Machine Learning.

Por primera vez, el anuncio de los ganadores fue hecho en el Campus São Paulo, un espacio de Google dedicado a los emprendedores. De este modo, la compañía intenta promover una discusión entre los académicos, la comunidad de startupsy directivos de universidades sobre uno de los grandes motores para la innovación en el que la región, tiene amplio espacio para mejorar: el acercamiento entre la academia y la comunidad de emprendedores.

Los ganadores de los Premios LARA 2017 y sus instituciones son:

Marcos André Gonçalves

Clebson C. A. de Sá

Universidade Federal de Minas Gerais, Brasil

Optimizing Ensembles of Boosted Additive Bagged Trees for Learning-to-Rank

The goal is to optimize a ranked list of documents related to specific information needs by training a model with documents already defined as relevant by specialists  using a “Learning-to-Rank”  process based on a combination of Machine Learning techniques.

Pedro Olmo Stancioli Vaz de Melo

Túlio Corrêa Loures

Universidade Federal de Minas Gerais, Brasil

Discussion-Based Entity Representation

The goal is to develop a method for extracting relevant information through comments from discussions made online. Eventually will create a tool that could create a summary about comments from the same topic, and group them with other related subjects.

Jorge Arigony-Neto

Guilherme Tomaschewski Netto

Universidade Federal do Rio Grande, Brasil

Low-cost autonomous stations for measuring the impacts of climate change on glaciers

The goal is to record glacier weather parameters and improve the resolution of glacier melt measurements through low-cost sensors.

Marcus Ritt

Alex Gliesch

Universidade Federal do Rio Grande do Sul, Brasil

Heuristic algorithms for fair land distribution and districting problems

The objective is to develop a computational method for dividing parcels of lands into smaller lots considering geographical, political and equitable qualities.

José Correa

Raimundo Sanoa

Universidad de Chile, Chile

Adaptive and Personalized Sequential Posted Prices

The goal is to design welfare maximizing mechanisms in an online setting. Such mechanisms should make irrevocable decisions when faced to a customer. Therefore, to achieve good performance, they should incorporate two basic ingredients: personalization meaning that different customers may be treated differently; and adaptivity meaning that the treatment of a customer may depend on the acquired knowledge until that point.

Gonzalo Navarro

Patricio Huepe

Universidad de Chile, Chile

Engineering Compressed Random Access Memories

This proposal aims at creating a layer between programs and the RAM memory they use so that all the stored data is automatically compressed in a way that is transparent to the program. In this way, the actual memory offered by the device is virtually expanded. While such a layer may introduce some performace overhead, we expect it to be much lower than the one introduced by virtual memory systems, which resort to external memory. Further, it can be used in devices with no external memory, like cellphones, sensors, and other low-end devices.

Edgar Emmanuel Vallejo Clemente

Kevin Islas Abud

Tecnológico de Monterrey, México

Predicting Zika epidemics using social and vectorial contact networks (project extension)

The main goal is to develop a method that could predict outbreaks and infections of the Zika virus through human-mosquito simulated interactions.

Mirko Zimic

Jorge Coronel

Universidad Peruana Cayetano Heredia, Peru

Improvements to facilitate the diagnostics of tuberculosis in low resources settings using mobile technologies and artificial intelligence

This project aims to improve a faster detection of tuberculosis (within 2 or 3 days) through a computational system which could have an online data analysis system to aid the diagnosis.

Luis Carlos González Gurrola

Ricardo Manuel Carlos Loya

Universidad Autónoma de Chihuahua, México

Learning Roadway Surface Disruptions patterns to improve transportation Infrastructure

This project objective is to develop a system for detecting road anomalies through drivers’ smartphones sensors and GPS. Also, with machine learning, classify and rank the data of the roads based on size, severity, damage, time unattended, etc. With this, correspondent offices can focus their efforts, and resources more efficiently.

Moacir Ponti

Patricia Bet

Universidade de São Paulo, Brasil

Mobile inertial sensors for fall risk screening and prediction

This project aims to do a research about the use of movement sensors from mobile devices for fall risk prediction in the elderly. The developed methodology could be implemented in mobile platforms such as smartphones or smartwatches. Sensors represent a viable monitoring option aiding carers and health professionals.

Fernando Magno Quintão Pereira

Junio C. Ribeiro da Silva

Universidade Federal de Minas Gerais, Brasil

Intelligent DVFS

This project aims to create a prototype for reducing energy consumption of Android applications at a minimum price in program’s performance. This prototype will use reinforcement learning to adapt itself to the different ways in which those applications can be used.

Juan Pablo Galeotti

Iván Arcuschin Moreno

Universidad de Buenos Aires, Argentina

Evolutiz: Multi-objective Test Generation for Testing Evolving Android Applications

This project aims to develop an open source tool for testing new Android applications, helping developers to identify errors or missing functionality. In addition, it intends to create a publicly available database of defects in Android applications.

Anderson Rocha

José Ramon Trindade Pires

Universidade Estadual de Campinas, Brasil

Automated Data­-Driven Screening of Diabetic Retinopathy – Extension

In this research, we aim to design methods to recognize discriminative patterns of diabetic retinopathy stages, providing an advanced and robust severity decision; and incorporate such information into a final higher-level (and more refined) decision of referable DR. We also intend to explore possible forms of understanding the decisions taken by the devised solutions toward accountable decision-making methods.

Pablo Arbelaez

Andres Romero

Universidad de los Andes, Colômbia

Learning Dynamic Action Units for Three-dimensional Facial Expression Recognition

Automated understanding of facial expressions is a fundamental step towards high-level human-computer interaction. This project plans to model human facial expressions through the analysis of temporal variations in the pattern of activations of three-dimensional Action Units. The automated facial expression analysis opens the door to multiple application domains beyond emotion classification, such as pain, drowsiness, intoxication, facial unspoken language, depression, or lie detection.

Rodrigo Coelho Barros

Jônatas Wehrmann

Pontifícia Universidade Católica do Rio Grande do Sul, Brasil

Order Embeddings and Character-level Convolutions for Multimodal Retrieval and Synopsis Generation

After a year of research, the team is now able to retrieve images that semantically match a query written in natural language in a very efficient way, as well as retrieving pre-stored descriptions given a query image. For the second year, the project aims to make use of this approach for being capable of retrieving videos given textual descriptions, as well as automatically describing them in natural language.

César Rennó-Costa

Ana Claudia Costa da Silva

Universidade Federal do Rio Grande do Norte, Brasil

Bringing sleep research into the realm of machine learning: optimization of a biologically-sound deep learning SLAM algorithm through offline self-organization of vertices

This project is inspired on the biology of sleeping as an activity of self-organization, aiming to apply it to machine learning improvements. Sleeping can optimize deep learning networks.

Cristina Nader Vasconcelos

Felipe Moure Cícero

Universidade Federal Fluminense, Brasil

Skin lesion classification, segmentation and dermoscopic patterns detection using deep learning

This project aims to develop new tools of skin image analysis which could make automated detection and classification (as a benign or malignant) of a skin lesion which would aid a medical diagnosis for early detection of skin cancer.

Eduardo Alves do Valle Junior

Michel Fornaciali

Universidade Estadual de Campinas, Brasil

Reliable Automated Melanoma Screening for the Real World

This project aims to speed-up real-world adoption of computer-aided melanoma screening, both by enhancing the Machine Learning models used to detect the disease, and by interacting with doctors to identify and remove barriers that prevent the adoption of the technology.

Hiram Eredín Ponce Espinosa

José Guillermo González Mora

Universidad Panamericana, Mexico

Transfer Learning Using Artificial Hydrocarbon Networks: A Case Study in Robotics

This research project aims to design a strategy using previous knowledge to solve new but similar problems quickly and effectively (transfer learning) based on artificial hydrocarbon networks. It will be implemented on robots for task learning. The research project is part of an ongoing research for developing a biomechanical robot for rescuing.

Teodiano Freire Bastos-Filho

Alexandre Luís Cardoso Bissoli

Universidade Federal do Espírito Santo, Brasil

Multimodal Assistive Domotics Including Augmentative and Alternative Communication

Last year, this project developed a new assistive system to be used by people with severe motor disabilities making them able to control home devices and communicate with people around or through a smartphone right from his/her wheelchair and the signals captured from muscles or eyes. In the current year, volunteers will take home the system to test. To make this possible, this year will be configured the system for each volunteer, and proposed a new device controller that will transmit the information to the home devices by internet.

Rodrigo F. Cádiz

Andrés Aparicio

Pontificia Universidad Católica de Chile, Chile

Auditory graphs: conveying data through sound for the visually-impaired

This project’s main goal is to convey quantitative information contained in visual graphs via sound, a technique called sonification, and when particularly applied to graphs, known as auditory graphs. This approach would allow blind or visually-impaired people to access and comprehend this information, otherwise only available in a visual form. This approach has many advantages over the standard method of tactile-graphs, as they are easy to implement with general-purpose computer equipment available today for educational purposes. Of particular interest to us is the development of tools to be used along with widespread web technologies such as images search engines or spreadsheets. We would like to add the capability of not only seeing the data using these tools, but also to hear it.

Anselmo Frizera-Neto

Andrés Alberto Ramírez-Duque

Universidade Federal do Espírito Santo, Brasil

Multimodal Interaction Environment based on Computer Vision and Robotic Device for Assisting the Diagnosis of Children with Autism Spectrum Disorder

Although there is no cure for autism spectrum disorder (ASD), intensive and early intervention is crucial for increasing child’s level of functioning in daily life. Many research groups are working on technology-based diagnosis and intervention tools for ASD. Nevertheless, the real benefits of their use in systematic clinical practices are still inconclusive. This research aims at developing an open technology-based tool to support clinicians in the diagnosis and therapeutic interventions for children with ASD. A smart room composed of a robotic device and a multi-camera computer vision setup for analyzing the behaviors of children with ASD is currently under development. In this manner, the researchers seek the development of tools to aid clinicians identify risk factors in children with ASD and provide resources for the next-generation of therapeutic interventions. This work is developed by a multidisciplinary team composed of clinical professionals (neuro-pediatricians and psychologists) and experts in biomedical engineering.

Wagner Meira Junior

Roberto C. S. N. P. Souza

Universidade Federal de Minas Gerais, Brasil

Hot Spot Mining from Case-Control Trajectories

The goal is to detect hotspots, that is, regions where the chance of occurrence of a target event (e.g., being infected by a disease) is higher compared to the rest of the area under analysis, based on trajectories recorded by personal devices.

Winston S. Percybrooks

Pedro Juan Narvaez Rosado

Universidad del Norte, Colômbia

Towards large scale, intelligent, computer-aided auscultation for remote primary-care settings

This proposal aims to build a low-cost, mobile phone-based digital stethoscope capable of recording and labeling heart sounds. It also looks to use Machine Learning to develop a method for automatic diagnosis of heart-related conditions from the recorded sounds. We expect the system would be suitable for use in telemedicine scenarios, particularly to support non-specialized medical practitioners at remote locations.

Sidarta Ribeiro

Ana Raquel Torres

Universidade Federal do Rio Grande do Norte, Brasil

Non-semantic graph analysis for automated assessment and early diagnosis of cognitive disabilities in the school environment

The goal consist in creating a low-cost automated assessment of cognitive disabilities that impair a full academic development of young students, using tools from natural language processing to contribute early detection and adequate intervention.

Maria da Graça Campos Pimentel

Raiza Tamae Feminino Hanada

University of São Paulo, Brasil

A Dwell-Free Eye typing intelligent tool for motor disabilities users

This project has already developed models able to suggest possible eye-typed words from severe motor disabilities users. This year is intended to improve the interface, perform experiments with potential real users, create datasets and deliver a low-cost open source software tool.

André R. A. Grégio

Fabricio Ceschin

Universidade Federal do Paraná, Brasil

Identifying Concept-Drift in Malware Classifiers and the Applicability of Deep-Learning Detectors for APTs

In this project, we intend to develop novel, adaptive models to identify whether a program is benign or malicious during its execution, even when its behavior is subtle, creating a real-time infection detector. We also intend to create a public dataset holding as many features from malicious programs as possible, which will benefit from our proposed techniques to be always up-to-date. The models will rely on deep learning for multi-stage classification, as well as innovative concept-drift detection techniques.

Compartir esta noticia