The multidisciplinary nature of this research area allows involved researchers and students to acquire a solid and broad training due to the applicability in several domains and the knowledge of the various facets of the addressed problems.
Data Analytics and Applications
The Data Management and Applications area encompass both the aspects related to the data structure, data storage, and location as well as the process of producing information and extracting knowledge from the data. Research in this area involves data structures and algorithms to support the data knowledge extraction process, as well as in the application and development of functions for preprocessing, machine learning and post-processing of data. In this area, it is appropriate to establish a datacentric treatment for those processes which, commonly, correspond to in-silico experiments.
This research area is dependent on the addressed problem domain and has a strong multidisciplinary character. The application of this research area are commonly observed in the government, scientific and business areas. In the government area, one can highlight applications in the areas of planning, health, transportation, education, and communication. In the scientific area, applications in astronomy, seismic and drugs stand out. In the business area, we highlight the development of software for the Information Technology market and its applications in marketing, social networks, multimedia systems, and smart cities. All these domains bring with them different technologies present in the data production and consumption area that can be exemplified from the Internet of Things (IoT), sensor networks to interactive applications.
In this flood of data, applications particular features and demands bring numerous challenges in managing large volumes of data. Regarding large volumes of data processing, there is also a pressing need to use high-performance processing (PAD) to achieve large-scale data analysis. There are significant challenges in the establishment of these analyzes, usually modeled as workflows. In these workflows, activities, and data are directed to execution in some PAD environment (e.g., clusters, grids, clouds).
The different applications, while presenting themselves as applied research, often provide the opportunity to elaborate new theoretical frameworks in basic research, of a more general character, for the solution of these practical problems. Basic research occurs both in establishing representations that are agnostic to the medium in which they will be performed, as well as in the elaboration of new functions that are more appropriate to the different data sources and that have a broad spectrum of use.
Machine Learning and Optimization
Much of the advances in Data Science related to extracting knowledge from data have been achieved through the application of machine learning and optimization techniques. Activities in this research project involve the development and application of machine learning and optimization techniques, as well as designing and using efficient algorithms for the analysis, understanding, forecasting, and description of complex problems in several domains, such as Biology, Computer Science, Engineering, Finances, Health, Logistics, among others.
Optimization is an area of expertise that works on modeling and developing efficient algorithms for solving both fundamental and applied optimization problems. Regarding the development of efficient optimization algorithms, one may classify them as either Exact, Heuristic or Hybrid. When talking about exact methods, common approaches are based on Linear Programming, Integer Programming, and Non-linear Optimization. As for Heuristics, Artificial Intelligence techniques and Metaheuristics are the usual approaches. At last, Hybrid methods consists in combining Exact, Heuristic, Data Mining, and High-Performance Computing techniques, in order to bypass difficulties of one of the previously explained classes of methods.
Machine Learning (ML) is a subarea of Artificial Intelligence that investigates how machines can learn some tasks once being provided with a set of experiences (in the form of historical data). Two subareas of ML are of particular interest to our research group: Deep Learning and Reinforcement Learning. Deep Learning encompasses a set of techniques devised to simulate human brain-behavior in tasks such as visual recognition, speech recognition, and natural language processing. These techniques attempt to produce high-level hierarchical representations of the input data, through layers of sequential processing. Reinforcement learning is concerned with how software agents can rationally execute actions in an environment in order to maximize their reward. RL can be applied to several tasks, such as planning, scheduling, games, and more.