Kunal Pratap Singh

I am an incoming Ph.D. student at the VILAB at EPFL this fall, where I'm advised by Prof. Amir Zamir. I work on Computer Vision, Embodied AI, and Reinforcement Learning.

Before grad school, I spent two amazing years in the PRIOR team at the Allen Institute for AI (AI2). Previously I was a research assistant at the Computer Vision Lab at GIST working with Prof. Jonghyun Choi. Even earlier, I finished my undergraduate studies at the Indian Institute of Technology, Roorkee.

Email  /  CV  /  Google Scholar  /  Semantic Scholar  /  Twitter  /  Github

I am interested in building actionable representations that enable embodied agents to physically reason about their environments and perform tasks for and alongside humans.

We propose the Scene Graph Contrastive Loss, an auxiliary objective that encourages the agent's belief to align its representation with a rich graphical encoding of its environments. We show results on Object Navigation, Multi-Object Navigation and ArmPointNavigation.

We propose Ask4Help, a framework that endows embodied agents with the ability to request expert help. We applying this framework to existing off-the-shelf Embodied-AI models and improve task performance on Object Navigation and Room Rearrangement.

We factorize the policy and perception into separate streams to train effective instruction following agents on the ALFRED benchmark.

We develop the first architecture search method for binary neural networks. We propose a new search space and cell design, and discover architectures that outperform floating point backbones.

Proposed a flexible Deep Learning based approach for exposure fusion and HDR Imaging from arbitrary number of exposure bracketed shots.

This template is stolen from here.