Kinjawl Bhattacharyya

Hi! I’m currently a predoctoral fellow at the Kotak IISc AI/ML Centre, working in the Representation Learning Lab at the Department of Electrical Communication Engineering, IISc Bangalore, under the supervision of Dr. Prathosh AP. My current research focuses on improving zero-shot adversarial robustness in vision-language models (VLMs) using test-time prompt tuning strategies. I’ve also recently contributed to work on domain generalisation in medical imaging using Langevin-based data augmentation, which is currently under review.

I completed my B.Tech. in Electrical Engineering from IIT Kharagpur, during which I had the opportunity to work at the Representation Learning Lab, IISc on multiple occasions. My earlier work involved unsupervised domain adaptation in medical imaging using energy-based models (EBMs) for image-to-image translation, which led to a co-authored publication in Elsevier’s Medical Image Analysis. Later, I explored topics around multi-concept compositionality and catastrophic neglect in diffusion-based text-to-image generation, working with latent consistency models (LCMs) and attention regularisation techniques at inference time.

In the summer of 2023, I interned at the Vision and Learning Lab at the University of Alberta with Dr. Li Cheng, supported by the Mitacs GRI Fellowship. There, I worked on building a multi-modal human motion dataset, including multi-sensor RGB-D calibration and explored monocular SLAM for eye-tracking pose estimation. I’m broadly interested in problems at the intersection of representation learning, robustness, and multimodal intelligence, and am currently looking for research assistant positions or long-term internships at machine learning and visual computing research groups. If my work aligns with your interests, feel free to check out my CV and don’t hesitate to reach out via email at kinjawl03@gmail.com or LinkedIn.