Focus on building compute- and data-efficient deep learning models for computer vision.

Associate Professor II working on efficient deep learning and computer vision problems.

Developed an unsupervised machine learning based automated event detection techniques to detect energy events in microseismics.

Over the course of stay at TU Delft, I had actively been involved in the following projects: 1. Developing an efficient high-resolution topology optimization (TO) method.– Proposed a combined mesh and penalization adaptivity based TO scheme.– Established uniqueness bounds and identified numerical artefacts in existing multiresolution TO methods.– Proposed a dp-adaptive scheme for high-resolution structural designs.2. Optimization of metallization design in solar cells.– Developed a novel 2D finite element model for numerical modeling of the electrical parameters.– Optimized metallization designs for freeform geometries and variable conditions.3. Deep learning based models for noise reduction and data recovery.– Designing a rock physics supermodel using stacked neural networks (NN).– Developing a stacked autoassociative NN framework to learn geophysical models for denoising purposes.

This project involved adding rotation utilities to GPlates software and processing and interpretation of ASTER data to understand the drifting of northern and southern Australia.

The internship focused on automated correlation of well logs and identification of lithological features from well data. As a part of this work, the milestones achieved were:– Formulation of a pattern-recognition approach based on sliding window and normalized cross-correlation.– Implementation of the algorithm using C# and Ocean API as a plugin for Petrel software.