My Research Journey so far....

Embarking on my research journey in May 2017 during my second year of MTech, I worked under the supervision of Dr. Mridu Sahu to pioneer an impactful Electroencephalogram (EEG) based Brain-Computer Interface (BCI) system. This innovation is geared towards enhancing the quality of life for neurologically disabled patients. We focused on making better EEG-based BCI systems, such as motor imagery-based BCI and P300 speller BCI, with the help of various machine-learning techniques. In 2018-19, I was involved in designing a speech recognition system using a convolution neural network and long short-term memory. This research, conducted in the Department of School Education, National Informatics Center, addresses the intricacies of English pronunciation. In 2021, I joined the Deep Intelligence Lab, Department of Computer Science and Engineering, IIT Indore, as a Junior Research Fellow under the guidance of Dr. Puneet Gupta. I contributed to the project entitled Heart rate monitoring from non-contact face videos using Deep Learning. This work focused on estimating and monitoring heart rate by utilising remote PhotoPlethysmoGraphy (rPPG), wherein the human's cardiac activity is monitored using the non-contact face videos acquired from camera sensors. From this project, I was able to publish papers at two conferences, WACV and ICPR. In PhD, my focus lies within the realms of deep learning, computer vision, and medical imaging. We aim to design a robust, trustworthy and better-performing medical image segmentation model. In our initial effort, we seek to assess the impact of adversarial attacks on this model's efficacy. To achieve this, we introduce a dynamic loss selection-based adversarial attack, MedIS Attack, incorporating non-differentiability issues. Our second proposed method, TrustMIS, investigates the trustworthiness of the medical image segmentation models using the consistency between input and rotated variants and improves the performance of non-trustworthy prediction. TrustMIS also considers the multiple models and selects the best one that provides the most trustworthy prediction. Currently, we are working on detecting anomaly samples (adversarial and OOD) and defending against adversarial attacks in order to enhance the robustness of medical imaging models.

To explore more about my research work and publications, kindly visit Research page and Publications page, respectively.