“My intellectual interest has long been in developing mathematical approaches to squeezing essential information from complex, noisy, real-world data.”
When asked to pinpoint her field of research, Erzsébet Merényi, a research professor in statistics and in electrical and computer engineering at Rice, avoids the phrase “big data” as vague and vogueish. “The real challenge for identifying significant knowledge from data is in the complexity,” she said.
Merényi earned her master’s degree in mathematics and Ph.D. in computational science from Szeged University in her native Hungary, and describes herself as a data analyst. Her research focuses on brain-like self-organized neural computation for manifold learning, pattern recognition, clustering, classification, variable selection and other aspects of high-dimensional data with complex structure.
“The brain funnels immense amounts of data to optimally summarized and organized representations on the cerebral cortex, and refines them by continual learning,” she said. “This enables fast and precise recognition of complex patterns, including discovery of the small and unusual.”
Merényi turned to brain-like information processing in the early 90’s after encountering the “jump” in complexity seen in hyperspectral imagery—millions of data samples, each with hundreds of variables. Traditional methods failed to fully exploit the intricate features recorded by advanced sensors, aimed at capturing relevant physical and chemical processes.
Tracing one theme—astronomy and space science—suggests the broad applicability of her research. In a Soviet-Hungarian-French-U.S. collaboration, it was Merényi’s mathematical restorations of severely degraded, once-in-a-lifetime spacecraft images from a 1986 rendezvous with Comet Halley that enabled the first three-dimensional kinematic model of a comet nucleus. Her comprehensive neural classifications of the 1997 panoramic spectral imagery from the Mars Pathfinder revealed Martian geologic trends that eluded conventional approaches.
Merényi’s work helped develop a new asteroid taxonomy and aided in the prediction of water in asteroids. Her hyperspectral applications characterized the surfaces of Mars, the Moon, Earth and Pluto. Ongoing collaboration with the Methodist Hospital utilizes similar techniques to produce detailed brain maps from fMRI data.
Most recently Merényi has started working with “ultra-spectral” data cubes from ALMA (Atacama Large Millimeter Array), the most advanced radio astronomy telescope array on the planet, in operation since 2013. It is a single telescope composed of 66 high-precision antennas.
“This is the most powerful radio-telescope in the world. With it we hope to answer some of the biggest astrophysical questions, from star and planet formation to the formation of the universe. Big data,” she said.