One of the fundamental goals of computer vision is to understand a scene. Towards this goal, we want the system to answer several questions - who, what, when, why, how much, etc. pertaining to the visual scene. As a graduate student my research focused on problems trying to answer simple questions like ‘who’ and ‘how much’. This involved the problems of re-identifying persons over a network of cameras and automatically summarizing large videos. As a postdoctoral researcher, I am working on complex questions like ‘what’ and ‘when’ for a visual scene. A challenging application scenario related to this is detecting activities in untrimmed videos. A step further into this problem is to describe what is going on in a video scene in natural language. Conducting research in these areas allows me to build on my previous experience on image/video analysis and provide opportunities to broaden my exposure to deep, end-to-end systems. Despite their enormous success, current deep neural networks (DNNs) are black boxes that do not expose their decision making process or whether they can be trusted and/or corrected. My future research will focus on the ‘why’ aspect of the DNNs to make them explainable and thus, more compatible with human reasoning.
Principal Investigator
- Making Unsupervised Domain Adaptation More Efficient
Ph. D. Students
Anurag Roy
Area of Research: Machine Learning and Information Retrieval
Kaushik Dey
Area of Research: Reinforcement Learning
Omprakash Chakraborty
Area of Research: Explainable AI
Owais Iqbal
Area of Research: Few shot learning
