Divyansh Jhunjhunwala

Hi! I am Divyansh, a third year PhD candidate in the Electrical and Computer Engineering department at Carnegie Mellon University, advised by Dr. Gauri Joshi. My research interests lie broadly in distributed optimization and machine learning, in particular federated learning.

In summer 22 and summer 23, I interned at IBM Research working with Dr. Shiqiang Wang on some interesting problems in federated learning. Recently I have also been interested in the problem of machine unlearning, where the goal is to make a model 'forget' part of the training data.

Prior to CMU, I completed my Bachelors in Technology (B.Tech) in Electronics and Electrical Communication Engineering from IIT Kharagpur, where I received the Institute Silver Medal for graduating with the highest CGPA in my department.

Email  /  CV  /  Google Scholar

May 23: I am returning to IBM Research, Yorktown Heights as a summer research intern.

Jan 23: My internship work on tuning the server step size in federated learning was accepted as a spotlight presentation at ICLR 2023!

Oct 22: Our work on incentivizing clients for federated learning was accepted as an oral presentation at the FL-Neurips 22 workshop! (12% acceptance rate).

Aug 22: Completed my internship at IBM T.J. Watson Research Center, New York.

April 22: Our team was selected as a finalist for the Qualcomm Innovation Fellowship for the research proposal "Incentivized Federated Learning for Data-Heterogeneous and Resource-Constrained Clients".

We present FedExP, a method to adaptively determine the server step size in FL based on dynamically varying pseudo-gradients throughout the FL process.

Propose IncFL algorithm to explicitly maximize the fraction of clients that are incentivized to use the global model in federated learning.

Propose FedVARP algorithm to deal with variance caused by only a few clients participating in every round of federated training.

Introduce notions of spatial and temporal correlations and show how they can be used to efficiently compute the mean of a set of vectors in a communication-limited setting.

Propose an adaptive quantization strategy that aims to achieve communication efficiency as well as a low error floor by changing the number of quantization levels during training in federated learning.

Source code credit to Dr. Jon Barron.