About me
I am a postdoc advised by Prof. Sriram Sankararaman at the University of California, Los Angeles. I am developing deep learning based methdologies to impute missing phenotype data in the UK biobank dataset. I obtained my Ph.D./M.A. under the supervision of Prof. Niraj K. Jha in Electrical and Computer Engineering at Princeton University. Earlier, I obtained M.S. Electrical Engineering from the University of Michigan, Ann Arbor where I worked with Prof. Heath Hofmann. I received my undergraduate in Electrical Engineering from Indian Institute of Technology (IIT), Gandhinagar. After my M.S., I worked as a systems engineer in the automotive industry.
My research interests involve the use of machine learning (ML) to design better and safer systems. My ML methods present an explainable decision framework and is agnostic to the amount of uncertainty. Here is a summary of my research during my Ph.D.
Inverse design for constrained multi-objective optimization (INFORM). A two-step optimization process. The first optional step is a modified genetic algorithm. The individuals in a generation are a mix of those generated by crossover and mutation and the ones injected using inverse design. In the second step, we select a reference solution and use inverse design to improve its performance further. See here for a short 3 min. video about INFORM.
Completing a partially-specified system: Given a system with incomplete inputs (some inputs known) and outputs (multi-dimensional), complete the rest of the inputs and the output.
Optimization of a partially-specified system: Similar to above, but in an active learning setting. A part of the system inputs are fixed and a designer is provided with a reference solution. The goal is to improve the system performance by searching over only the variable inputs.
Anomaly detection/location/correction: Given an observed system behavior (multi-dimensional inputs and response), identify if the observation is correct. If an anomaly is detected, identify the features (can be input or response) that are incorrect, and correct the incorrect features.
Uncertainty modeling: When a system has the same response for the different sets of input, determine all possible input combinations that yield the same response and assign confidence to each input.
I have experience working in varied fields like power electronics, power systems, motor control, besides my current research in machine learning.
Feel free to contact me to learn more about my research or possible collaboration opportunities.
Copyright © 2023, Prerit Terway