Sagnik Majumder

Hi, all! Welcome to my webpage. I hope you find some resources that might be of use to you.

I am currently a first year Master of Science in Computer Science (MS CS) student at the University of Texas at Austin (UT Austin). Previously, I was serving as a research assistant at the Center for Computation and Cognition (CCC), Department of Computer Science and Mathematics, Goethe University in Frankfurt under the supervision of Prof. Visvanathan Ramesh and Mr. Martin Mundt. I was broadly working on deep learning for computer vision at that time. Even before that, I had spent four wonderful years at the Birla Institute of Technology and Science Pilani (BITS Pilani) before graduating with a Bachelor of Engineering (Hons) in Electronics and Instrumentation in July 2018.

My areas of research have included investigating algorithms for better and more efficient machine learning (ML) that can be practically applied to diverse computer vision tasks in parallel or in tandem. Such systems require mechanisms for the retainment and transfer of old learned knowledge. They offer a tremendous potential for transfer and continual learning research. Confidence bounds estimation through Bayesian inference has also been an integral part of my work which not only helps induce robusteness in a continual learning framework but also makes it more data efficient. I have also extensively researched in the field of automated model search with reinforcement learning. I believe integrating model search and efficient knowledge transfer algorthims is a way to realize robust and efficient self-adapting continual learning systems. Further, I have gained some experience of building human-brain-inspired neural models of vision that have the ability to learn through mutual excitations and inhibitions of neural connections while having access to limited training data (read ‘not big data!’) through my time with Prof. Christoph Malsburg at the Frankfurt Institute for Advanced Studies (FIAS).

I have a few interesting research areas in mind which I would love to work on in the future, apart from my current/previous areas of work. Some of these are few-shot learning, adversarial defense and embodied learning. In my view, these fields still hold a lot of scope for improvement and my previous experience might allow me to address the problems with some fresh perspective.

Broad research areas

• Meta-learning and continual learning
• Deep reinforcement learning
• Bayesian inference
• Deep learning in computer vision
• Principled cognitive models of perception