I am a Senior Research Scientist with IBM's Almaden Research Center in San Jose, CA. My current research is centered around large language models (LLMs), with a focus on training data quality, privacy, and safety. Before LLMs, my research was focused on on building mathematical foundations of machine learning to address privacy, security, and scalability challenges.

Prior to joining IBM Research, I was a postdoctoral researcher at the EECS Department of University of California, Berkeley, where I am working with Kannan Ramchandran. During my postdoc, I designed and analyzed scalable algorithms for privacy-preserving and robust distributed machine learning and for enabling distributed computing on emerging platforms such as serverless systems and blockchains. In my research, I applied fundamental ideas from information and coding theory, signal processing, optimization, and cryptography in a principled and foundational manner. 

Prior to joining UC Berkeley, I completed my PhD from the ECE Department of Texas A&M University, where I was advised by Alex Sprintson. In my doctoral research, I developed coding-theoretic techniques for cloud storage systems to provide security against eavesdropping attacks and ensure high data availability for fast content download. I also designed efficient algorithms for private information retrieval that enable users to access records from a remote database while preserving the privacy of queries.

I have been extremely fortunate to work with Emina Soljanin (Rutgers University),  Robert Calderbank (Duke University), and Sidharth Jaggi (University of Bristol) during my PhD and postdoc.

Before starting my PhD, I worked as an R&D Engineer at the Innovation Labs of TATA Consultancy Services, Bangalore. The theme of my work was “Signal Processing and Coding Techniques for Futuristic Communications and Storage”, with a focus on compressive sensing and network coding.

I obtained my master's degree from the Electrical Engineering Department of Indian Institute of Technology Kanpur in July 2009. My thesis was on “Multi-antenna Broadcast Channels with Limited Feedback”, advised by A. K. Chaturvedi.

News

• (Feb 2025) 'GneissWeb' - a new 10T token dataset for pre-training LLMs (Gneiss is a metamorphic rock, pronounced as nice)

  • IBM Research Blogpost, Hugging Face GneissWeb Page

• (Oct 2024) 'Protecting Users From Themselves: Safeguarding Contextual Privacy in Interactions with Conversational Agents' in NeurIPS 2024 Workshop on Socially Responsible Language Modeling

• (Jun 2024) 'Split, Unlearn, Merge: Leveraging Data Attributes for More Effective Unlearning in LLMs' in ICML 2024 FM-Wild Workshop

• (Dec 2023) 'FairSISA: Ensemble Post-Processing to Improve Fairness of Unlearning in LLMs' in NeurIPS 2023 workshop on Socially Responsible Language Modeling

• (Dec 2023) 'Data Poisoning Attacks on Generative Models' paper in NeurIPS 2023 workshop

• (Oct 2023)  Extended paper on Privacy-Preserving Federated Learning associated with our prize winning entry at the US PETs Prize Challenge is on arXiv

• (Apr 2023) Paper on Communication Efficient Vertical Federated Learning in ICML 2023

• (Sep 2022) Our team won a prize in  the US PETs Prize Challenge

• (July 2022) Paper on Federated Unlearning in ICML 2022 workshop

• (May 2022) Our team won a prize in MICCAI Federated Tumor Segmentation Competition. Paper is on arXiv.

• (Sep 2021) Paper accepted in NeurIPS 2021

• (Sep 2021) INSPIRE Lab Seminar at Rutgers University on Privacy-Preserving Federated Learning (video)

• (Sep 2021) ACCESS Seminar on FastShare

Old News...

• (Feb 2021) Invited talk at Facebook

• (Nov 2020) Gave a tutorial at 2020 European School of Information Theory

• (Nov 2020) Invited talk at Intel Labs

• (Oct 2020) OverSketched Newton paper accepted in IEEE BigData 2020

• (Sep 2020) Invited talk at Google Research

• (Sep 2020) FastSecAgg paper accepted in CCS Workshop on Privacy-Preserving Machine Learning in Practice

• (Sep 2020) Paper accepted in IEEE Conference on Blockchain

  • CoVer: Collaborative Light-Node-Only Verification and Data Availability for Blockchains

• (June 2020) Paper accepted in ICML Workshop on Federated Learning for User Privacy and Data Confidentiality

  • FastSecAgg: Scalable Secure Aggregation for Privacy-Preserving Federated Learning 

• (Apr 2020) Two papers at International Symposium on Information Theory 2020

  • Communication-Efficient Gradient Coding for Straggler Mitigation in Distributed Learning

  • Communication-Efficient and Byzantine-Robust Distributed Learning

• (Feb 2020) Invited Talk at Information Theory and Applications (ITA) Workshop

• (Dec 2019) Our paper on Analyzing the Download Time of Availability Codes is now on arXiv

• (Sep 2019) Presented paper at Scaling Bitcoin in Tel Aviv

  • SeF: A Secure Fountain Architecture for Slashing Storage Costs in Blockchains

• (Aug 2019) I will be participating in the Proofs, Consensus, and Decentralizing Society program at Simons Institute

• (Aug 2019) Presented paper at Information Theory Workshop in Visby

  • On an Equivalence Between Single-Server PIR with Side Information and Locally Recoverable Codes

• (Jul 2019) Three papers at International Symposium on Information Theory (ISIT) in Paris

  • Gradient Coding Based on Block Designs for Mitigating Adversarial Stragglers

  • Low-degree Pseudo-Boolean Function Recovery Using Codes

  • Single-Server Multi-Message Individually-Private Information Retrieval with Side Information

• (Jul 2019) Paper on arXiv: Logical Clifford Synthesis for Stabilizer Codes

• (Jun 2019) Paper in Coding Theory for Large-Scale Machine Learning (CodML) workshop at ICML

  • OverSketched Newton: Fast Convex Optimization for Serverless Systems (Spotlight)

• (Feb 2019) Invited talk at Information Theory and Applications (ITA) Workshop

• (Sep 2018) Paper is accepted in Allerton 2018

• (May 2018) Two papers are accepted in ISIT 2018

• (Apr 2018) Attended workshop on Coding and Information Theory at CMSA, Harvard

• (Mar 2018) Presented our work on binary codes with availability at AMS Sectional Meeting 2018

• (Feb 2018) Presented our work on rank and subspace locality at ITA 2018

• (Jan 2018) Started postdoc at UC, Berkeley

• (Sep 2017) Papers got accepted in 'Allerton 2017'

• (Apr 2017) Papers got accepted in 'ISIT 2017'

• (Jan 2017) Received an invitation to give a graduation day talk at 'ITA 2017'

• (Sep 2016) Presented paper at 'Allerton 2016'

• (July 2016) Presented paper at 'ISIT 2016' at UPF, Barcelona, Spain 

• (Jun 2016) Attended 'North American School of Information Theory' at Duke University

• (Jun 2016) Attended 'Algorithmic Coding Theory' workshop at Brown University

• (Apr 2016) Paper on Codes with Unequal Locality got accepted in ISIT

• (Feb 2016) Attended amazing series of workshops at Institute of Henry Poincare Paris, France "Nexus 2016"

• (Jan 2016) Paper uploaded on ArXiv (Codes with Unequal Locality)

• (Dec 2015) Attended excellent DIMACS workshop on "Network Coding: the Next 15 Years"

• (Oct 2015) Paper in Allerton

• (Sep 2015) Got invitation to attend Banff Workshop on Mathematical Coding Theory for Multimedia Streaming 

• (Apr 2015) Papers accepted in ISIT

• (Jan 2015) Paper accepted in ITW

• (Dec 2014) Abstract got selected for Joint Math Meeting 2015 to be held at San Antonio

• (Oct 2014) Giving a seminar in TAMU ISS Seminar series ("Coding for the Clouds: An Overview of Novel Erasure Codes for Distributed Storage")

• (Oct 2014) Got selected for SP Coding School to be held at Brazil from Jan 18-31, 2015

• (July 2014) Gave a talk at Bell Labs ("Reliable, Deniable, and Hidable Communication over Multi-path Networks" )

• (June 2014) Gave a talk at University of Toronto ("Information-Theoretic Security for Communication and Storage: Looking Beyond the Perfect Secrecy")

• (May 2014) Received NSF scholarship to attend North American School of Information Theory (NASIT) at Toronto, Canada  

• (Apr 2014) Paper accepted in NetCod

• (Apr 2014) Paper accepted in ISIT

• (Jul 2013) Started internship at Chinese University of Hong Kong