Publications
Rec-Distill: An Industrial Distillation Pipeline for Large-Scale Recommendation Models
Haoran Ding, Wenlin Zhao, Yuchen Jiang, Juren Li, Jie Zhu, Xinchun Li, Yishujie Zhao, Yi Zhang, Ao Qiao, Jianhui Dong, Cheng Chen, Ziyan Gong, Deping Xie, Peng Xu, Zikai Wang, Yuwei Wang, Huizhi Yang, Zhe Chen, and Yuchao Zheng
arXiv preprint, 2026
Rec-Distill bridges the gap between large recommendation models and latency-constrained online serving. It combines large-teacher scaling with decoupled training, black-box distillation, debiasing, and a hybrid batch-streaming pipeline for dynamic recommendation environments. The framework scales teachers to 24B dense parameters and 20K behavior sequences while allowing lightweight students to recover more than 60% of teacher gains in the best setting, with improvements also validated in online recommendation and advertising scenarios.
LiPM: Foundation Model for Lithium-Ion Battery Analysis
Juren Li, Yang Yang, Hanchen Su, Jiayu Liu, Youmin Chen, Jianfeng Zhang, and Lujia Pan
KDD, 2025
LiPM is a pretrained foundation model designed for heterogeneous lithium-ion battery datasets and irregular sampling protocols. It combines a mix-masked autoencoder for electrochemical consistency, a Coulombic Integration Regression objective that encodes charge conservation, and a dual-scale temporal encoder for local irregular timestamps and long-range dynamics. Pretraining across eight battery datasets enables transfer to different battery types, partial charge-discharge segments, and downstream analysis tasks.
Chromosomal Structural Abnormality Diagnosis by Homologous Similarity
Juren Li, Fanzhe Fu, Ran Wei, Yifei Sun, Zeyu Lai, Ning Song, Xin Chen, and Yang Yang
KDD, 2024
Structural chromosome abnormalities are difficult to identify because chromosome morphology varies and subtle defects require expert comparison. HomNet follows the diagnostic principle that normal homologous chromosomes should have matching structures: it adaptively aligns homologous pairs, models their differences, and aggregates evidence across multiple pairs to reduce noise and detect structural abnormalities on real clinical data.
Paper · Code · Chinese introduction · Video
Disentangling Domain and General Representations for Time Series Classification
Youmin Chen, Xinyu Yan, Yang Yang, Jianfeng Zhang, Jing Zhang, Lujia Pan, and Juren Li
IJCAI, 2024
CADT studies unsupervised domain adaptation for time-series classification by explicitly separating domain-invariant representations from domain-specific ones. A class-wise hypersphere objective improves the decision margin of the transferable representation, while domain-preserving augmentations help capture domain-specific patterns. The framework was evaluated on public datasets and multiple real-world applications.
DWLR: Domain Adaptation under Label Shift for Wearable Sensor
Juren Li, Yang Yang, Youmin Chen, Jianfeng Zhang, Zeyu Lai, and Lujia Pan
IJCAI, 2024
Wearable-sensor models face distribution shifts across users and devices, including changes in both features and class proportions. DWLR addresses these two shifts together through learnable label-distribution reweighting, information-gain regularization, and separate alignment in the time and frequency domains. Across three wearable-sensor datasets, the method improves average performance by 5.85% over prior approaches.
How Powerful are Interest Diffusion on Purchasing Prediction: A Case Study of Taocode
Xuanwen Huang, Yang Yang, Ziqiang Cheng, Shen Fan, Zhongyao Wang, Juren Li, Jun Zhang, and Jingmin Chen
SIGIR, 2021
This work studies purchase prediction from the perspective of information diffusion in Taocode, a product-sharing mechanism on Taobao. Based on more than 100 million real-world sharing records, it introduces InfNet, a dynamic graph neural network with structural and temporal modeling to capture how product interests spread between users and influence subsequent purchases.