Yijun Liu

PhD candidate, Electrical and Computer Engineering, USC (advisor Richard M. Leahy)

I build machine learning models that compress high-dimensional neural/behavioral signals into compact, transferable embeddings.

Previously, I earned my M.Eng. degree in ECE from Cornell and a B.Eng. degree in Communication Engineering from Soochow University.

Contact: yijunl AT usc DOT edu

HSTU-BLaIR: Lightweight Contrastive Text Embedding for Generative Recommender
Workshop on Large Language Models for E-Commerce, KDD 2025 [pdf][code]
sEEG-based Encoding for Sentence Retrieval: A Contrastive Learning Approach to Brain-Language Alignment
Workshop on What is Next in Multimodal Foundation Models, CVPR 2025 [pdf]
Untamed: Unconstrained Tensor Decomposition and Graph Node Embedding for Cortical Parcellation
Y. Liu, J. Li, J.L. Wisnowski, R.M. Leahy.
bioRxiv 2025 [project page] • expanded from oral at BioKDD 2023 [slides]
Identification of Overlapping and Interacting Networks Reveals Intrinsic Spatiotemporal Organization of the Human Brain
J. Li, Y. Liu, J.L. Wisnowski, R.M. Leahy.
Neuroimage, 2023 [pdf]
A Hybrid High-Resolution Anatomical MRI Atlas with Sub-Parcellation of Cortical Gyri using resting fMRI
A.A. Joshi, S. Choi, Y. Liu, M. Chong, G. Sonkar, J. Gonzalez-Martinez, D. Nair, J.L. Wisnowski, J.P. Haldar, D.W. Shattuck, H. Damasio, R.M. Leahy.
Journal of Neuroscience Methods, 2022 [pdf]
Brain Network Decomposition for Naturalistic Stimulus Paradigm
Y. Liu, J. Li, J.L. Wisnowski, A.A. Joshi, R.M. Leahy.
Organization for Human Brain Mapping Annual Meeting (OHBM), 2021 [pdf]
Pancreas Segmentation in Abdominal CT Scans
Y. Liu, S. Liu.
International Symposium on Biomedical Imaging (ISBI), 2018 [pdf][code]

Bronze Medal (Top 8%), Optiver “Trading at the Close” price movement prediction competition, 2024
Solo submission among 4,436 teams using LightGBM ensemble with lagged volatility and order book features.
[certificate]
Discovered and reported a bug in MATLAB’s silhouette function for clustering evaluation using the correlation metric.
Identified output deviations from the documented formula and scikit-learn; confirmed by MathWorks and fixed in R2021b.
[MathWorks confirmation email]
Reported and debated a behavioral mismatch in SciPy’s csgraph.laplacian for normalized Laplacians.
Initiated discussion around handling of self-loops, leading to long-term reconsideration by core developers.
[GitHub issue #14490]
2nd Place, Columbia Health Hackathon, 2018
Team of 4 across 4 countries. Developed a Flask-based web app for pancreas segmentation using deep learning.
[project link]