Yash Maurya
Privacy-focused research engineer in AI and differential privacy
Summary
Work
Education
Projects
Writing
When Privacy Guarantees Meet Pre-Trained LLMs: A Case Study in Synthetic Data
July 1, 2025Conference presentation and case study (USENIX PEPR '25) examining privacy challenges when differential privacy meets pre-trained LLMs, including empirical findings about synthetic data leakage.
AI Governance and Accountability: An Analysis of Anthropic's Claude
May 1, 2024Analysis of AI governance and accountability frameworks applied to Anthropic's Claude, identifying threats and proposing mitigation strategies.
Unified Locational Differential Privacy Framework
May 1, 2024A unified framework for locational differential privacy enabling private aggregation of various data types over geographical regions using local DP mechanisms.
Guardrail Baselines for Unlearning in LLMs
January 1, 2024Paper showing that simple guardrail approaches (prompting, filtering) can match finetuning for unlearning in LLMs and highlighting limitations of current benchmarks.
Position: LLM Unlearning Benchmarks are Weak Measures of Progress
January 1, 2024Position paper arguing that existing LLM unlearning benchmarks can give overly optimistic views of unlearning effectiveness and proposing recommendations for future evaluations.
Federated Learning for Colorectal Cancer Prediction
January 1, 2022IEEE conference paper demonstrating federated learning for privacy-preserving colorectal cancer prediction across hospitals.