Prateek Yadav

Continual Learning, Generalization, Sparsity, MoE, Parameter Efficiency, Graphs.

praty@cs.unc.edu

Hey! I am a PhD student at the MURGe-Lab at the University of North Carolina - Chapel Hill, where I work with Prof. Mohit Bansal. My research goal is to make deep learning models learn continually and generalize to multiple domains. I am interested in efficient methods that lead to generalization and exploiting sparsity, memory, and mixture-of-expert models for continual learning.

Previously, I have worked on a diverse set of topics – 1) Interpretability, 2) compositional reasoning in NLP, 3) deep learning methods for Graph and Hypergraph structured data and their application to NLP, 4) estimated and controlled for uncertanity in the learned representations from these methods, and 5) worked on Bayesian modeling of temporal data.

Over the past few years, I have been fortunate to work with Ming Tan, Qing Sun, Xiaopeng Li at Amazon AWS AI Labs, with Prof. Partha Talukdar at MALL-Lab at Indian Institute of Science (IISc) Bangalore, with Dr. Prateek Jain at Microsoft Research India and with Prof. Arun Rajkumar at Indian Institute of Technology, Madras. I also worked for a year with some amazing people at LinkedIn AI Bangalore. Before all this, I completed my undergraduate degree in pure mathematics in 2018 from IISc Bangalore where I was supervised by Prof. Partha Talukdar.

Check out my publication page to know more about my current research. I am also actively involved in improving the way science is taught to students, through simple experiments and conducting workshops via Notebook Drive.

News

Nov 2022 Serving on the program committe of Graphs and More Complex Structures for Learning and Reasoning at AAAI 2023.
Oct 2022 I have been awarded the Amazon Science - Post Internship Fellowship!
Oct 2022 Pre-print for our paper Exclusive Supermask Subnetwork Training for Continual Learning is now available!
Oct 2022 Serving on the program committe of Transfer Learning for NLP Workshop 2022 at NeurIPS 2022.
Apr 2022 New paper on Contrastive Learning for Explanation Graph Generation is accepted at ACL 2022, Dublin, Ireland.
Feb 2022 Pre-print for our paper Low-Cost Algorithmic Recourse for Users With Uncertain Cost Functions is now available!
Aug 2021 ExplaGraphs: An Explanation Graph Generation Task for Structured Commonsense Reasoning is accepted at EMNLP 2021, Punta Cana, Dominican Republic.
Mar 2021 multiPRover: Generating Multiple Proofs for Improved Interpretability in Rule Reasoning is accepted at NAACL 2021.
Jan 2021 Pre-print for our paper Discrete Time Latent Hawkes Processes for Modeling Multidimensional Temporal Event Streams is now available!
Nov 2020 Our paper Rank Refinement: An Algorithmic framework with applications to diversity aware influence maximization has been accepted at GCLR workshop at AAAI 2020.