My main research interest lies with inverse problems in acoustic imaging. Specifically, medical ultraosund and geophysics. I work on techniques for physics informed unsupervised deep learning to solve inverse problems in real time, volumetric imaging, analytic inversion methods, and non-linear signal processing.
While inverse problem approaches such as full waveform inversion (FWI) and travel time tomography have been prevalent in seismic imaging for several decades, they are still a subject of active research. These methods tend to be sensitive to initial conditions, have difficulty with piecewise smooth domains, are slow, computationaly intensive, and still often require a human in the loop interaction.
Deep learning based methods allow real time results and are not dependent on initial conditions. Regularization can be injected via appropriate training. Unfortunately though, real world training sets are virtually non-existent and limitations due to domain transfer call for unsupervised domain aware training to allow training networks on real unclassified data.
Interleaving the two approaches allows enjoying the blessings of both worlds with results benefitting more than just the ultrasound and seismic domains.
I currently hold a position as a research scientist at the Department of Mechanical engineering as well as the Institute for Medical Engineering and Science (IMES) at MIT
I have a background both in industry and academia working on inverse problems, deep learning, perturbation methods, sparsity, optical time of flight, RADAR, acoustic imaging, medical image processing, signal processing, as well as GPU computing and HPC.
Massachusetts Institute of Technology
PhD in Applied Mathematics
M.Sc in Applied Mathematics
B.Sc in Mathematics and Computer Science