Yash Maurya

Graduate Student
School of Computer Science
Carnegie Mellon University


Email: ymaurya [at] cs.cmu.edu
LinkedIn: yashmaurya
GitHub: yashmaurya01
Scholar: yashmaurya01

I am a graduate student in the Privacy Engineering Program at Carnegie Mellon University, pursuing my passion for creating privacy-conscious AI solutions and ensuring the ethical use of data. My mission is to design robust privacy systems for the greater good of society.

Currently, I am working as a Research Assistant in the School of Computer Science, sponsored by PwC, under the guidance of Professor Norman Sadeh, Professor Lorrie Cranor, and Professor Hana Habib. We are building a User-Centric Privacy Notice and Choice Threat Modeling Framework. I'm also working with Professor Virginia Smith and Professor Steven Wu to explore Guardrails for Unlearning in Large Language Models (LLMs).

Before joining CMU, I worked as an R&D Engineer at Samsung, where I primarily contributed to prototyping Samsung News' new Recommendation System. Prior to that, I also worked as a Federated Learning Researcher at DynamoFL(YC W22)

My key research interests encompass Privacy Preserving Machine Learning, Fairness, Federated Learning, Differential Privacy, Explainable and Responsible AI, and ML Safety. Currently, I am actively working on Unlearning in LLMs, Privacy Threat Modeling, and Implicit Bias Auditing.

If you'd like to chat, please feel free to reach out at ymaurya [at] cs.cmu.edu

For more details, please see Resume


NEWS

March 2024

Received the Project Olympus Spark Grant by Swartz Center for Entrepreneurship for building a Language Learning agent to help non-native speakers improve their speaking skills
LinkedIn Post

       

Our work - 'Through the Lens of LLMs: Unveiling Differential Privacy Challenges' got accepted at PEPR(Privacy Engineering Practice and Respect) 2024 USENIX Conference

       

Jan 2024

I completed the Certified Information Privacy Technologist (CIPT) credential from the IAPP - International Association of Privacy Professionals!
Credential

Sept 2023

Won Space Theme in HackCMU'23 with our Space-JEDI Project
LinkedIn Post DevPost


Research

March 2024

Through the Lens of LLMs: Unveiling Differential Privacy Challenges

Aman Priyanshu, Yash Maurya, Suriya Ganesh, Vy Tran
2024 USENIX Conference on Privacy Engineering Practice and Respect (PEPR'24)
PEPR Link

March 2024

Guardrail Baselines for Unlearning in Large Language Models

Pratiksha Thaker, Yash Maurya, Virginia Smith
Secure and Trustworthy Large Language Models - ICLR 2024 Workshop(SeT LLM @ ICLR 2024)
arXiv

December 2023

Is it worth storing historical gradients to identify targeted attacks in Federated Learning?

Joong Ho Choi, Yingxin Liu, Yash Maurya
PDF GitHub

December 2022

Federated Learning for Colorectal Cancer Prediction

Yash Maurya, Prahaladh Chandrahasan
IEEE Global Conference for Advancement in Technology 2022 (IEEE GCAT'22)
IEEE Link

December 2022

Improved variants of Score-CAM via Smoothing and Integrating

Rakshit Naidu, Haofan Wang, Soumya Snigdha Kundu, Ankita Ghosh, Yash Maurya, Shamanth R Nayak K, Joy Michael
Responsible Computer Vision - CVPR 2021 Workshop (RCV @ CVPR 2021)
RCV Poster

October 2020

IS-CAM: Integrated Score-CAM for axiomatic-based explanations

Rakshit Naidu, Ankita Ghosh, Yash Maurya, Shamanth R Nayak K, Soumya Snigdha Kundu
arXiv


Projects


Feb 2024

BiasBusterDPGen
Prompt-driven synthetic data augmentation using counterfacuals for bias correction with differential privacy alternative
GitHub


Sept 2024

Space-JEDI (Junk Elimination and Debris Interception)
Space-JEDI is a predictive software that utilizes real-time NASA data to track satellites, forecast their future positions, and generate optimal flight plans for space debris collection vehicles, enabling effective monitoring and management of objects in Earth's orbit.
GitHub