Home

I work as a Research Scientist at Meta. Previously, I was a Senior Applied Scientist in the Responsible AI team of Amazon AGI group. I graduated from the PhD program at the Department of Computer Science of USC in 2019, where I was working with Prof. Shrikanth Narayanan.

Recent Updates

• Two of our papers were accepted at the ACL 2026 main conference: From Narrow Unlearning to Emergent Misalignment: Causes, Consequences, and Containment in LLMs and SWAN: Semantic Watermarking with Abstract Meaning Representation
• I'm excited to join the EMNLP 2026 organizing committee as one of the Ethics co-chairs.
• I'm serving as an Area Chair at ARR March 2026 cycle.
• Our paper 'Constrained Entropic Unlearning: A Primal-Dual Framework for Large Language Models' was accepted at NeurIPS 2025.
• Two papers on LLM Unlearning (LUME, Selective Unlearning) accepted to EMNLP 2025 Findings.
• Starting a new role at Meta as a Senior Research Scientist.
• I will be serving as an Ethics chair at EMNLP 2025.
• Our paper 'Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond' was accepted at ICML 2025.
• New paper 'Towards Safety Reasoning in LLMs: AI-agentic Deliberation for Policy-embedded CoT Data Creation' was accepted at ACL 2025.
• New paper titled 'REAL Sampling: Boosting Factuality and Diversity of Open-Ended Generation via Asymptotic Entropy' to appear in Transactions of Association for Computational Linguistics (TACL), 2025.
• 2 papers accepted at AAAI 2025 workshop on Preventing and Detecting LLM Misinformation.
• Concluded LLM unlearning challenge at SemEval.
• 3 papers with collaborators accepted at NAACL 2025.
• Amazon Nova family of models launched!
• Our benchmark paper for indirect prompt injection attacks accepted at Neurips AdvML-Frontiers 2024 workshop.
• Our paper on improving contrastive decoding by extrapolating probabilities was accepted at EMNLP 2024 (with Best Paper Nomination from a Reviewer).
• LLM unlearning challenge accepted at SemEval 2024.
• New paper titled 'Tree-of-Traversals: A Zero-Shot Reasoning Algorithm for Augmenting Black-box Language Models with Knowledge Graphs' was accepted to main conference at ACL 2024.