Research
Machine-learning research spanning biomedical modeling, LLM safety, and NLP.
FedCSIS · published in IEEE
@inproceedings{liang2025stacking,
author = {Liang, Alan and Liu, Cenzhi and Wang, Kai and Liu, Ethan},
title = {A Stacking-Based Ensemble Approach for Predicting Chess Puzzle Difficulty},
booktitle = {Proceedings of the 20th Conference on Computer Science and Intelligence Systems (FedCSIS)},
series = {Annals of Computer Science and Information Systems},
volume = {43},
pages = {819--824},
year = {2025},
publisher = {Polish Information Processing Society},
doi = {10.15439/2025F1698},
issn = {2300-5963}
}Under review at NeurIPS
Center on AI Research for Health (AI4Health), USC Information Sciences Institute · Los Angeles, CA
Benchmarked TCR–pMHC binding affinity prediction methods against an in-house dataset, then prototyped a new approach for cancer-vaccine research.
University of California, Riverside · Riverside, CA
Measured what unlearning costs a language model in diversity, fluency, and utility, using BLEURT to catch meaning that degraded.
Modeling uncertainty in immunology data
Lead project · USC Information Sciences Institute
Testing whether positive-unlabeled learning can separate true non-binding TCR–pMHC pairs from ones that were never tested.
Fairness in NLP
Course research project · CSCI 567 (Machine Learning)
Implemented adversarial debiasing and group-specific threshold tuning on a BERT toxicity classifier to reduce AAE/SAE false-positive-rate disparities.
Repository-level code LLMs
Course research project · CSCI 544 (Natural Language Processing)
Built retrieval baselines on RepoBench and tested whether hallucination-aware context selection beats similarity-based retrieval for multi-file code completion.