EECS500 Fall 2016 Department Colloquium

Yi-Qiao Song
Porous Media, Magnetic Resonance and Machine Learning
Schlumberger-Doll Research
White 411
October 20, 2016
Porous material is ubiquitous in nature and human life. Rocks, soil, concrete, wood, food, and biological tissues are good examples. They are intrinsically multi-phasic, and their microstructure is critical for their functions. In recent years, NMR/MRI has become an important technique for characterization of a variety of porous media for petroleum exploration, material sciences, and medical imaging. This talk will outline a wide range of techniques used for in situ measurement of the material porosity, and their physical mechanisms. Examples of NMR/MRI methods and applications will be discussed including multi-echo techniques, compressed sensing, 2d methods for diffusion and relaxation to study complex diffusion dynamics in porous media and their applications in the study of polymer degradation, molecular composition, porosity in sedimentary rocks, food and biological tissues.
In particular, we will highlight the challenges these techniques face in the practical implementation and analysis of data with limited signal-to-noise. We will discuss Bayesian approach to understand the uncertainty of the data analysis and the machine learning concept for realtime optimization of the data acquisition to achieve fast and robust measurements.
Dr Yiqiao Song is a scientific advisor at Schlumberger-Doll Research in Cambridge MA and also works part-time at Martinos Center of Biomedical Imaging of Massachusetts General Hospital. His research involves development of nuclear magnetic resonance and imaging techniques and instrumentation to understand complex materials and fluids. His interest focuses on the physics of diffusion dynamics in porous media and biological tissues and the development of multi-dimensional experimental methods and numerical inversion algorithms. These multi-dimensional experiments have been broadly used in research and industrial applications. One of his current areas of interest is the Bayesian theory, uncertainty, and machine learning as a means to optimize NMR/MRI data acquisition in realtime in order to enhance the speed and quality of the experiments and for robust/automated applications. He is a fellow of American Physical Society, and a member of the Editorial Board of Journal of Magnetic Resonance.