About Me
Hello world! I am Ishan Khurjekar.
Update: I am a now a postdoctoral fellow at the Johns Hopkins University !! (new website)
Over the years, I have worked on multiple projects with 1 aspect in common: sensing. Sensing methods are the mainstay of scientific and engineering innovation. For example, sensing methods are used for locating submarines in the sea or damages in structures or even imaging objects which are invisible to the eye.
I enjoy researching and developing methods to understand, quantify, and manage the effects of uncertainty in various sensing-based engineering / scientific applications. My research goal is to enhance the reliability and utility of machine learning (ML) methods for such applications. My past + current work has focused on these topics:
- Uncertainty quantification: Develop techniques to assess the reliability of ML methods when external uncertainty sources are present in the data.
- Uncertainty management: Develop ML methods that are robust to external variations present during data collection.
- Utility enhancement: Obtain interpretable insights from ML models trained on sensing data to elucidate the underlying complex physical processes.
I am open to new collaborations and am generally excited to talk about anything research-related!
Uncertainty quantification
1. Principled UQ for inverse problems in acoustical signal processing [JASA'23, ASA'23, IEEE MLSP'23]: Slides
We consider the acoustic inverse problem of direction-of-arrival (DOA) estimation. We want to provide statistically valid and easy-to-obtain uncertainty intervals for the DOA estimate in the presence of external uncertainty. Conformal prediction (CP) technique is suited for this task. CP takes in predictions and produces an interval containing the true value with a statistical guarantee. We demonstrate the usage of CP with multiple machine-learning models trained for DOA estimation. CP is able to produce uncertainty intervals that are statistically valid and can reliably assess the DOA estimation performance in the presence of uncertainty.
Uncertainty management
1. Reliability assessment for structural health monitoring [NDT&I submitted]: Slides
We consider the task of assessing the reliability of a guided-wave structural health monitoring system. In a structural monitoring setup, sensors transmit and receive guided waves. Based on the received signals, we want to localize the damage. The challenge is to assess the reliability of localization predictions when external uncertainties are present. We analyze the ability of multiple deep-learning methods to represent the reliability of individual damage localization estimates in the presence of external uncertainty. We are able to obtain a heuristic uncertainty estimate in a simple and scalable manner.
2. Uncertainty management for structural health monitoring [JASA'22, ICASSP'22]: Slides
Here we demonstrate the ability of specialized machine learning techniques to generate robust predictions in the presence of external sources of uncertainty. We show that the ensemble of deep neural networks trained solely on simulated observations with adversarial perturbations provides improvements in damage detection and localization performance over baseline methods.
Enhancing utility
1. Obtaining interpretable insights for grain growth [Frontiers'23, QNDE'22]:
Grain growth in polycrystalline materials is a complicated physical process. Application of external heat or pressure results in a particular grain orientation (texture). Identifying the texture property is crucial to understanding the grain growth mechanism. Grain images obtained from a sensing setup can be used for analysis.
We demonstrate the usage of convolutional neural networks to classify grain images from a scanning electron microscope (sensing setup) in an automated manner. Further the intermediate features of the convolutional network provide potential insights about the grain growth mechanism.
LET'S CHAT
