Multiscale visual analytics - The solution to understand big data

MLG-INGI seminar

Event Speaker: Alexandru Telea
Event Place: Paul Otlet Seminar Room
Event Date: 11/09/2015 - 12:0

The last decade has witnessed unprecedented developments in the size, complexity, variability, and availability of data produced by scientific simulations, measurements, and sensing devices. Making full sense of such 'big data' spaces has become an increasingly critical problem in many application areas, ranging from scientific modeling to software maintenance, medicine, engineering, and geosciences. Gathering so-called actionable insight, or facts and patterns deeply embedded in large amounts of heterogeneous, multivariate, noisy, and time-dependent data requires efforts that combine diverse disciplines such as data mining and machine learning, sampling and signal theory, and interactive data visualization.

In recent years, visual analytics has become a key component of the data analytics toolset. Visual analytics combines scientific and information visualization techniques (for the interactive exploration of large data spaces) with data mining techniques (for the automatic detection of patterns, trends, correlations, and outliers in data) and machine learning tools (for high-level tasks such as classification, recognition, and categorization). In this talk, we present several recent research directions that make visual analytics an efficient, scalable, and versatile technique for data exploration: We show how image-based computational approaches, combined with GPU-based computation models, make visual analytics able to handle datasets of millions of points and hundreds of dimensions amenable to multiscale interactive visual exploration; how classical automated data mining and pattern recognition can be seamlessly integrated with user-driven exploration for making sense of complex phenomena embedded in the data; and how the above procedures can lead to the development of efficient and effective end-user tools supporting various application domains. We illustrate this talk with examples of big data analysis from software maintenance, large-scale graph analysis, 2D and 3D shape processing, and medical imaging.