STAR-Bench: Probing Deep Spatio-Temporal Reasoning as Audio 4D Intelligence [arXiv]

Zihan Liu^*, Zhikang Niu^*, Qiuyang Xiao, Zhisheng Zheng, Ruoqi Yuan, Yuhang Zang, Yuhang Cao, Xiaoyi Dong, Jianze Liang, Xie Chen, Leilei Sun, Dahua Lin, Jiaqi Wang

We formalize audio 4D intelligence, defined as reasoning over sound dynamics across time and 3D space, and introduce STAR-Bench to measure it, with a focus on linguistically hard-to-describe acoustic cues.

Homepage| Github |

SongGen: A Single Stage Auto-regressive Transformer for Text-to-Song Generation [ICML 2025]

Zihan Liu, Shuangrui Ding, Zhixiong Zhang, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Yuhang Cao, Dahua Lin, Jiaqi Wang

A single-stage auto-regressive transformer for text-to-song generation that offers versatile control via lyrics, descriptive text, and an optional reference voice while supporting both mixed and dual-track modes to meet diverse requirements

Homepage| Github |

SongComposer: A Large Language Model for Lyric and Melody Generation in Song Composition [ACL main 2025]

Shuangrui Ding^*, Zihan Liu^*, Xiaoyi Dong, Pan Zhang, Rui Qian, Junhao Huang, Conghui He, Dahua Lin, Jiaqi Wang

A language large model that understands and generates melodies and lyrics in symbolic song representations.

Homepage| Github ｜

An NCDE-based framework for universal representation learning of time series [IJCAI 2024]

Zihan Liu, Bowen Du, Junchen Ye, Xianqing Wen, Leilei Sun

An NCDE-based framework learns universal time-series representations via joint reconstruction and contrastive self-supervision, delivering strong performance across diverse downstream tasks and showing notable robustness to missing data.

Github ｜

Learning the evolutionary and multi-scale graph structure for multivariate time series forecasting [KDD 2022]

Junchen Ye^*, Zihan Liu^*, Bowen Du, Leilei Sun, Weimiao Li, Yanjie Fu, Hui Xiong

We propose an evolutionary and multi-scale graph learning framework that models dynamic dependencies among multivariate time series, achieving superior forecasting performance across domains such as transportation, energy, and finance.

Github ｜

Adaptive spatio-temporal graph neural network for traffic forecasting [KBS 2022]

Xuxiang Ta, Zihan Liu, Xiao Hu, Le Yu, Leilei Sun, Bowen Du

We propose a dynamic traffic graph structure with macro-level self-learning and micro-level self-adaptation, demonstrating strong interpretability and effectiveness in traffic forecasting.

Github ｜