Robin Jia
Assistant Professor of Computer Science
Studies robustness, interpretability and evaluation of NLP models and large language models.
Robin Jia's lab at USC works on interpretability, evaluation & safety of LMs, large language models and trustworthy & safe ML. 44 papers since 2023, 25 at top venues, mostly EMNLP and ACL. 7 PhD students. 10 alumni, most often to Microsoft. Applications close .
At a glance
- papers, 2023–2026
- 11/ year
- at T1 venues
- 57%
- PhD students
- 7
- MS students
- 0
- alumni listed
- 10
Papers by yearfrom the lab homepage, 2023–2026
- 14 papers2026 · T1 9 · collecting
- 12 papers2025 · T1 5
- 8 papers2024 · T1 5
- 10 papers2023 · T1 6
T1 venuesOtherVenues: EMNLP, ACL, ICLR, Findings of ACL, ICML
Research focuspaper titles that use each topic's words
- Large language models29 papers
- Interpretability, evaluation & safety of LMs4 papers
- interpretability3 papers
- evaluation1 paper
- Trustworthy & safe ML0 papers
Paper list
From the lab homepage.
202614 papers
- ICLR (T1)Cancer-Myth: Evaluating Large Language Models on Patient Questions with False Presuppositions
- COLMContextLeak: Auditing Leakage in Private In-Context Learning Methods
- COLMConvergent Evolution: How Different Language Models Learn Similar Number Representations
- ICML (T1)EPSVec: Efficient and Private Synthetic Data Generation via Dataset Vectors
- ICLR (T1)FoNE: Precise Single-Token Number Embeddings via Fourier Features
- ICLR (T1)Function Induction and Task Generalization: An Interpretability Study with Off-by-One Addition
- VL/HCCGenerating Complex Code Analyzers from Natural Language Questions
- ICML (T1)How Language Models Process Negation
- ICLR (T1)Hubble: a Model Suite to Advance the Study of LLM Memorization
- ICRA (T1)PSALM-V: Automating Symbolic Planning in Interactive Visual Environments with Large Language Models
- ACL (T1)Textual Steering Vectors Can Improve Visual Understanding in Multimodal Large Language Models
- COLMToken-Level Off-Policy Learning for Faithful Generation Under Distribution Shift
202512 papers
- FAccTInterrogating LLM Design under a Fair Learning Doctrine
- NAACL (T1)Language Models can Infer Action Semantics for Classical Planners from Environment Feedback
- COLMLLM Unlearning Without an Expert Curated Dataset
- Findings of ACLMechanistic Interpretability of Emotion Inference in Large Language Models
- EMNLP (T1)Promote, Suppress, Iterate: How Language Models Answer One-to-Many Factual Queries
- EMNLP (T1)Rethinking Backdoor Detection Evaluation for Language Models
- Findings of ACLRobust Data Watermarking in Language Models by Injecting Fictitious Knowledge
- COLMTeaching Models to Understand (but not Generate) High-risk Data
- ICLR (T1)TLDR: Token-Level Detective Reward Model for Large Vision Language Models
- Empirical Methods in Natural Language Processing: System DemonstrationsTokenSmith: Streamlining Data Editing, Search, and Inspection for Large-Scale Language Model Training and Interpretability
- Findings of ACLVerify with Caution: The Pitfalls of Relying on Imperfect Factuality Metrics
- EMNLP (T1)Why Do Some Inputs Break Low-Bit LLM Quantization?
20248 papers
- NAACL (T1)Do Localization Methods Actually Localize Memorized Data in LLMs?
- NAACL (T1)Efficient End-to-End Visual Document Understanding with Rationale Distillation
- COLMIsoBench: Benchmarking Multimodal Foundation Models on Isomorphic Representations
- AAAI/ACM AIESOperationalizing content moderation "accuracy" in the Digital Services Act
- NeurIPS (T1)Pre-trained Large Language Models Use Fourier Features to Compute Addition
- Findings of ACLProving membership in LLM pretraining data via data watermarks
- NeurIPS (T1)Transformers Learn Higher-Order Optimization Methods for In-Context Learning: A Study with Linear Models
- EMNLP (T1)When Parts are Greater Than Sums: Individual LLM Components Can Outperform Full Models
202310 papers
- ACL (T1)Are Sample-Efficient NLP Models More Robust?
- Findings of EACLBenchmarking Long-tail Generalization with Likelihood Splits
- EMNLP (T1)Chain-of-Questions Training with Latent Answers for Robust Multistep Question Answering
- ACL (T1)Contrastive Novelty-Augmented Learning: Anticipating Outliers with Large Language Models
- ACL (T1)Data Curation Alone Can Stabilize In-context Learning
- ACL (T1)Do Question Answering Modeling Improvements Hold Across Benchmarks?
- O-DRUMDoes VLN Pretraining Work with Nonsensical or Irrelevant Instructions?
- Findings of EMNLPEstimating Large Language Model Capabilities without Labeled Test Data
- Findings of EMNLPHow Predictable Are Large Language Model Capabilities? A Case Study on BIG-bench
- EMNLP (T1)SCENE: Self-Labeled Counterfactuals for Extrapolating to Negative Examples
Where alumni went10 listed, 20% to industry
Industry 1Academia 4
Most often to
- Microsoft
Also to
- UC Berkeley
- Stanford University
- USC
Apply to USCFall 2027
Computer Science PhD (Viterbi)
- Round
- Single round (Dec 15)
- Opens
- —
- GRE
- See program page
- English
- See program page
- Fee
- See program page
Confirmed on the program page.
Similar labsnearest by research summary
- CNPeking
Ge Li
- Large language models
- USUSC
Swabha Swayamdipta
- Interpretability, evaluation & safety of LMs
- CAToronto
Zhijing Jin
- Large language models
- USNYU
He He
- Interpretability, evaluation & safety of LMs
- USUCSD
Taylor Berg-Kirkpatrick
- Large language models
- GBImperial
Marek Rei
- Language & society
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
USCcloses