Mengyue Wang

Xinjin Li, Jinghan Cao, Mengyue Wang et al.

Traffic forecasting requires modeling complex temporal dynamics and long-range spatial dependencies over large sensor networks. Existing methods typically face a trade-off between expressiveness and efficiency: Transformer-based models capture global dependencies well but suffer from quadratic complexity, while recent selective state-space models are computationally efficient yet less effective at modeling spatial interactions in graph-structured traffic data. We propose FAST, a unified framework that combines attention and state-space modeling for scalable spatiotemporal traffic forecasting. FAST adopts a Temporal-Spatial-Temporal architecture, where temporal attention modules capture both short- and long-term temporal patterns, and a Mamba-based spatial module models long-range inter-sensor dependencies with linear complexity. To better represent heterogeneous traffic contexts, FAST further introduces a learnable multi-source spatiotemporal embedding that integrates historical traffic flow, temporal context, and node-level information, together with a multi-level skip prediction mechanism for hierarchical feature fusion. Experiments on PeMS04, PeMS07, and PeMS08 show that FAST consistently outperforms strong baselines from Transformer-, GNN-, attention-, and Mamba-based families. In particular, FAST achieves the best MAE and RMSE on all three benchmarks, with up to 4.3\% lower RMSE and 2.8\% lower MAE than the strongest baseline, demonstrating a favorable balance between accuracy, scalability, and generalization.

Xin Li, Mengyue Wang, T. -P. Liang

Recommender systems are a critical component of e-commercewebsites. The rapid development of online social networking services provides an opportunity to explore social networks together with information used in traditional recommender systems, such as customer demographics, product characteristics, and transactions. It also provides more applications for recommender systems. To tackle this social network-based recommendation problem, previous studies generally built trust models in light of the social influence theory. This study inspects a spectrumof social network theories to systematicallymodel themultiple facets of a social network and infer user preferences. In order to effectively make use of these heterogonous theories, we take a kernel-based machine learning paradigm, design and select kernels describing individual similarities according to social network theories, and employ a non-linear multiple kernel learning algorithm to combine the kernels into a unified model. This design also enables us to consider multiple theories' interactions in assessing individual behaviors. We evaluate our proposed approach on a real-world movie review data set. The experiments show that our approach provides more accurate recommendations than trust-based methods and the collaborative filtering approach. Further analysis shows that kernels derived from contagion theory and homophily theory contribute a larger portion of the model.

Mengyue Wang, Shuo Chen, Kristian Kersting et al.

Recent advances in Video Large Language Models (VLLMs) have significantly enhanced their ability to understand video content. Nonetheless, processing long videos remains challenging due to high computational demands and the redundancy present in the visual data. In this work, we propose METok, a training-free, Multi-stage Event-based Token compression framework designed to accelerate VLLMs' inference while preserving accuracy. METok progressively eliminates redundant visual tokens across three critical stages: (1) event-aware compression during vision encoding, (2) hierarchical token pruning in the prefilling stage based on semantic alignment and event importance, and (3) a decoding-stage KV Cache optimization that further reduces memory consumption. Our experiments on diverse video benchmarks demonstrate that METok achieves an optimal trade-off between efficiency and accuracy by dynamically selecting informative visual tokens. For instance, equipping LongVA-7B with METok realizes an 80.6% FLOPs reduction and 93.5% KV Cache memory savings, all while maintaining comparable or even superior accuracy.