Akshay Kumar
 |
|
About me
I'm currently pursuing my Ph.D. in the Electrical and Computer Engineering department at University of Minnesota Twin Cities, where I am advised by Jarvis Haupt. My research interests are in machine learning and optimization, particularly on deep learning theory. Before starting my Ph.D., I received my B.Tech.-M.Tech. (Dual Degree) in Electrical Engineering from Indian Institute of Technology Kanpur.
My recent research has focused on understanding the gradient flow dynamics of homogeneous neural networks trained in the small initialization regime. Our work uncovers phenomena such as directional convergence and the emergence of sparsity structures among the weights. These insights ultimately led us to a greedy algorithm for training deep neural networks, called Neuron Pursuit. I have also summarized my work in tweet1 and tweet2.
Publications
Learning Neural Networks by Neuron Pursuit
Akshay Kumar and Jarvis Haupt.
Preprint.Towards Understanding Gradient Flow Dynamics of Homogeneous Neural Networks Beyond the Origin
Akshay Kumar and Jarvis Haupt.
Accepted at JMLR.Early Directional Convergence in Deep Homogeneous Neural Networks for Small Initializations
Akshay Kumar and Jarvis Haupt.
Accepted at TMLR (Awarded J2C Certification).Directional Convergence Near Small Initializations and Saddles in Two-Homogeneous Neural Networks
Akshay Kumar and Jarvis Haupt.
Accepted at TMLR.Convexifying Sparse Interpolation With Infinitely Wide Neural Networks: An Atomic Norm Approach
Akshay Kumar and Jarvis Haupt.
IEEE Signal Processing Letters, 2020Group Sparsity via Implicit Regularization for MIMO Channel Estimation
Akshay Kumar, Akshay Malhotra and Shahab Hamidi-Rad.
IEEE Conference on Wireless Communications and Networking, 2023Hybrid Maximum Depth-kNN Method for Real Time Node Tracking using Multi-Sensor Data
Sudhir Kumar, Abhay Kumar, Akshay Kumar and Rajesh Hegde.
IEEE International Conference on Communications (ICC) 2015