Nilanjan Ray

Department of Computing Science
University of Alberta, Canada

I have extensively worked on image analysis and computer vision applications mostly with continuous optimization methods, graph algorithms and machine learning. My interest in deep learning for computer vision is more recent.

My latest research interest is ADEPT Computer Vision. I am taking the liberty of coining an acronym ADEPT: Architecture for Differentiable End-to-end Programming and Training. Differentiable programming refers to auto-differentiable computations in a processing pipeline. I am interested in an extension of it. ADEPT refers to end-to-end training/optimization performed by differentiable programming in an architecture that may consist of diverse data processing components. Some of the components may be non-differentiable, or, may have variable dimensions at the input and the output ends. My initial thoughts on ADEPT and my recent work within ADEPT are here. I am always looking for ways to work with fewer labeled images without compromising accuracy/performance. One of our ADEPT methods for left ventricle segmentation shows such characteristics.

The following list contains some of my past and present image analysis / computer vision applications and techniques.