barun.das [at] dbmi.emory.edu

Software Engineer, Department of Biomedical Informatics, School of Medicine, Emory University

I am a machine learning researcher interested in the practical applications of deep learning. I recently graduated from Georgia Tech with an MS in Computer Science, and I am looking to gain more real-world research experience. At Georgia Tech, I worked on a multimodal action recognition setting, where we combined vision and narration to improve activity recognition performance. This was done in a semi-supervised paradigm which eliminated the need for accurate, frame-level annotations. This meant that our method was significantly cheaper compared to traditional action recognition paradigms, where the datasets must be painstakingly annotated so that the model can learn from it. I also worked on other deep learning problems such as distribution shifts in domain generalization. We analyzed state-of-the-art methods and developed a new learning objective which showed impressive results in overcoming distributional shifts on various domain adaptation datasets. In addition to this, I have also worked on recommendation systems, federated learning, and knowledge graphs. While I have worked with computer vision the most, I am also keen to explore learning from other modalities such as language and sensor data. Besides research, I also have industry experience, having worked full-time at GE as a software engineer. Most recently, I worked as a software engineering intern at Autodesk in the summer of 2022, where I helped develop a graph-based recommendation system for their digital learning and certification platform.