The Ivanov regularization for sparse vectors is a popular option for enforcing sparse solutions for various machine learning and signal processing problems. Enforcing a vector to lie within a L1 ball of suitable radius is a popular approach, for which several efficient algorithms exist. In this talk we will consider a generalization of the L1 ball to a vector k-norm ball. We will discuss an algorithm to project onto the vector k-norm ball and suggest some ways to make it more efficient and fast in practice when dealing with high-dimensional vectors.
P. Balamurugan is an Assistant Professor at IEOR, IITB. He works on theoretical and practical aspects related to machine learning and deep learning problems.