Ketchup talk by Ishant (PhD_CSE student) on 08/08/2018
Title: Efficient inference algorithms for higher order MRF-MAP problems.
Time: 4:00 PM
Abstract: Many tasks in computer vision and machine learning can be modelled as the inference problems in an MRF-MAP formulation and can be reduced to minimizing a submodular function. Size of this objective function is the number of pixels in the image which can be million for the typical computer vision task. Running time complexity of traditional submodular function minimization is polynomial in number of pixel which makes them very slow for this task. In this talk, I will present the following frameworks for improving inference time by exploiting the underlying structure of the MRF.
1. A general framework of decomposing the inference problem into multiple smaller sub-problems.
2. A common framework for leveraging the complementary properties of two seemingly different methods of inference.