Research Scholar
Welcome to my website! I am Mayank Kumar, a PhD student at the Indian Institute of Technology Jodhpur, specializing in machine learning with a focus on federated learning. My research interests include handling statistical heterogeneity in distributed learning systems that can scale to real-world scenarios. On this website, you can explore my research, publications, and interests. I aim to share my knowledge and insights on various topics related to machine learning and federated learning, and I hope my website can be a valuable resource for anyone interested in these fields. Thank you for visiting! style="text-align: justify;"
2022
BAFL is a proposed approach to reduce communication costs in federated learning. Rather than compressing the model, BAFL fine-tunes only the top layers on local data and ablates base layers while transferring the model. This results in a reduction of communication costs without compromising accuracy. BAFL uses a model pre-trained on large-scale data as the global model to better initialize weights and reduce communication costs. Evaluations on VGG-16 and ResNet-50 models on WBC, FOOD-101, and CIFAR-10 datasets showed up to two orders of magnitude in communication cost reduction in both IID and Non-IID settings.
2019 - Present
Indian Institute of Technology, Jodhpur, RJ
2016 - 2018
National Institute of Technology, Patna
2012 - 2016
Poornima University, Jaipur, Rajasthan
2007 - 2009
Central Board of Secondary Education
Address
LAB 217A, CSE Department, Indian Institute of Technology Jodhpur
Phone
+91 9261444448
kundalwal.1@iitj.ac.in