Robustness & Evaluation
Stress-testing models and the benchmarks we judge them by, to surface where they really break.
UC Irvine · Natural Language Processing
Probing, explaining, and fixing how language models fail.
We are the Natural Language Processing group at UC Irvine. We study the robustness, interpretability, and reliability of machine learning and NLP — from benchmarks and explanations to neuro-symbolic reasoning and controllable generation.
Research
Interpretability & Explanations
Explaining what models do — and exposing the limits and manipulability of post-hoc explanations.
Reasoning & Controlled Generation
Bringing logical structure, constraints, and steering to neural models and LLMs.
Recent work
All publications →- Association for Computational Linguistics (ACL) · 2026 Lost in Simulation: LLM-Simulated Users are Unreliable Proxies for Human Users in Agentic Evaluations
- Association for Computational Linguistics (ACL) · 2025 Nudging: Inference-time Alignment of LLMs via Guided Decoding
- International Conference on Learning Representations (ICLR) · 2024 What's In My Big Data?