Tianci Liu

I am a Ph.D. Candidate in ECE@Purdue University under the supervision of Prof. Jing Gao. Before coming to Purdue, I spent two wonderful years at University of Michigan to acquire my MS degree in Statistics. Prior to that, I got my BS degree from Xiamen University.
My research goal is to develop principled methods for building knowledgeable and efficient machine learning models. My work is primarily focused on the following pillars:
-
Knowledgeable & Efficient LLMs: I design scalable methods for knowledge editing, retrieval-augmented generation (RAG), and efficient fine-tuning to build precise, adaptable, and resource-efficient (M)LLMs, enabling seamless integration of diverse knowledge sources in real-world deployments.
-
Trustworthy AI in a Data-Efficient Manner: I create principled methods to understand and improve fairness and integrity in AI systems with minimal data, mitigating risks and delivering reliable outcomes with minimal data requirement.
-
Applications: I apply AI/ML to real-world challenges across domains, such as sensing systems, education, biomedical imaging, and physics-informed modeling.
I am on the job market and am open to academic positions and industrial research roles. If you believe I might be a good fit for your institution or organization, please feel free to reach out to me liu3351[AT]purdue.edu
news
Aug 20, 2025 | Our paper “Towards Universal Debiasing for Language Models-based Tabular Data Generation” and “Learning to Instruct: Fine-Tuning a Task-Aware Instruction Optimizer for Black-Box LLMs” were accepted at EMNLP’25 Findings. |
---|---|
May 15, 2025 | Our paper “RoseRAG: Robust Retrieval-augmented Generation with Small-scale LLMs via Margin-aware Preference Optimization” was accepted at ACL’25 Findings. |
May 08, 2025 | Our paper “RAM-Hand: Robust Acoustic Multi-Hand Pose Reconstruction Using a Microphone Array” won Best Paper Award at Sensys’25. |
May 06, 2025 | Our paper “Beyond Invisibility: Learning Robust Visible Watermarks for Stronger Copyright Protection” was accepted at UAI’25. |
May 01, 2025 | Our paper “Mitigating Heterogeneous Token Overfitting in LLM Knowledge Editing” was accepted at ICML’25. |
selected publications
- ACL’25 FindingsRoseRAG: Robust Retrieval-augmented Generation with Small-scale LLMs via Margin-aware Preference OptimizationIn Findings of the Association for Computational Linguistics: ACL 2025, 2025
- ICML’25Mitigating Heterogeneous Token Overfitting in LLM Knowledge EditingIn The Fourty-Second International Conference on Machine Learning, 2025
- ICLR’25Unlocking Efficient, Scalable, and Continual Knowledge Editing with Basis-Level Representation Fine-TuningIn The Thirteenth International Conference on Learning Representations, 2025
- UAI’25Beyond Invisibility: Learning Robust Visible Watermarks for Stronger Copyright ProtectionIn The Fourty-First Conference on Uncertainty in Artificial Intelligence, 2025
- ICML’24LIDAO: Towards Limited Interventions for Debiasing (Large) Language ModelsIn The Fourty-First International Conference on Machine Learning, 2024
- ICLR’24Towards Poisoning Fair RepresentationsIn The Twelfth International Conference on Learning Representations, 2024
- AAAI’23Simfair: A unified framework for fairness-aware multi-label classificationIn Proceedings of the AAAI Conference on Artificial Intelligence, 2023
- ICML’23Optimization for amortized inverse problemsIn The Fortieth International Conference on Machine Learning, 2023
- EMNLP’24RoseLoRA: Row and Column-wise Sparse Low-rank Adaptation of Pre-trained Language Model for Knowledge Editing and Fine-tuningIn Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, 2024
- NeurIPS’24Counterfactual Fairness by Combining Factual and Counterfactual PredictionsIn The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024