arXiv subjects

Mengyue Wang

Publications and source records attributed to Mengyue Wang.

8 recordsLinked to original sources

arXiv2026

MOSAIC: Orchestrating Collaborative Knowledge Tracing with Hierarchical Semantic Alignment

Knowledge Tracing (KT) is important for personalized education but traditionally suffers from two key limitations: a reliance on shallow ID-based representations that neglect semantic depth and a restriction to single-granularity mastery estimation that overlooks hierarchical knowledge dependencies. To address these challenges, we propose MOSAIC (Multi-granularity Online Semantic AI for Collaborative Knowledge), a novel framework that orchestrates LLM-driven semantic alignment with sequential modeling. Unlike methods that use LLMs solely as predictors, MOSAIC leverages a frozen LLM to generate dynamic, context-aware embeddings and hierarchical prediction prompts, explicitly capturing collaborative signals and peer interactions. Furthermore, we introduce a cross-granularity consistency objective that jointly regularizes mastery estimation across concept, topic-cluster, and global proficiency levels. Extensive experiments on ASSISTments, EdNet, and a newly collected large-scale MOOC dataset demonstrate that MOSAIC establishes new state-of-the-art results. Specifically, our method achieves AUC improvements of up to 3.4\% and Accuracy gains of up to 2.5 \% across all benchmarks. Notably, MOSAIC exhibits superior robustness in collaboration-rich environments and long-sequence scenarios (AUC 0.862 on MOOC), offering both high predictive precision and semantically grounded interpretability.

cs.LG↗

arXiv2026

DeepLook: Deeper Thinking with Lookahead

Inference-time scaling has emerged as a powerful paradigm for improving large language model reasoning, often delivering larger gains on difficult reasoning tasks than parameter scaling alone. However, existing approaches remain inefficient in how compute is allocated within a reasoning trace. Motivated by the observation that reasoning failures often exhibit an early onset of uncertainty before a wrong answer become explicit, we introduce DeepLook, a training-free monitor-and-intervene decoding framework that concentrates lookahead compute at uncertainty bottlenecks. DeepLook aggregates token-level confidence into segment-level signals, triggers when confidence drops relative to recent history, and explores candidate continuations with fixed-horizon lookahead. Branches are ranked by Average Lookahead Confidence (ALC), the average segment-level confidence over rollout continuations, then pruned and aggregated through voting. On four competition-style mathematics benchmarks across DeepSeek-R1-8B, Qwen3-32B, GPT-OSS-20B, and GPT-OSS-120B, DeepLook shifts the accuracy--token-cost Pareto frontier: it improves accuracy over DeepConf-low in 11 of 16 settings while reducing dataset-level token generation by 87.3% on average, including gains of +3.1 on AIME25 with Qwen3-32B and +8.8 on BRUMO25 with GPT-OSS-20B. These results show that selective, future-aware intervention yields substantially stronger accuracy--cost trade-offs than uniformly scaling complete reasoning trajectories. Code is available here.

cs.AI↗

arXiv2026

FAST: A Synergistic Framework of Attention and State-space Models for Spatiotemporal Traffic Prediction

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.

cs.LG↗

arXiv2025

METok: Multi-Stage Event-based Token Compression for Efficient Long Video Understanding

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.

cs.CV↗

arXiv2025

Efficient defect healing of single-walled cabron nanotubes through $ \mathrm{C}_{2}\mathrm{H}_{2} $-assisted multiple-cycle treatment with air exposure

Defects in single-walled carbon nanotubes (SWCNTs) degrade their mechanical,electrical, and thermal properties, limiting their potential applications. To realize the diverse applications of SWCNTs, it is essential to enhance their crystallinity through effective defect healing. However, traditional thermal treatments typically require temperatures above 1800°C, which can alter the nanotube structure. Previously, defect healing of SWCNTs was achieved at a relatively low temperature of 1100°C, using C$_{2}$H$_{2}$ assistance, but the efficiency was limited. In this study, we developed a C$_{2}$H$_{2}$-assisted multiple-cycle process at an even lower temperature of 1000°C combined with air exposure, achieving highly efficient defect healing while preserving the nanotube structure. The combination of multiple-cycle treatment and air exposure between cycles was found to promote defect activation, suppress the formation of amorphous carbon, and enhance the effectiveness of defect healing. Additionally, we successfully healed commercially available bulk-scale SWCNTs (super-growth SWCNTs), noting that their healing behavior differed from lab-grown SWCNTs with smaller diameters synthesized from nanodiamond. The efficient and structure-preserved healing process developed in this study broadens the potential applications of high-quality SWCNTs, including flexible electronics, high-performance composites, and energy storage devices.

cond-mat.mtrl-sci↗

arXiv2024

A multi-theoretical kernel-based approach to social network-based recommendation

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.

cs.SI↗

arXiv2023

Gas flow-directed growth of aligned carbon nanotubes from nonmetallic seeds

Kite growth is a process that utilizes laminar gas flow in chemical vapor deposition to grow long, well-aligned carbon nanotubes (CNTs) for electronic application. This process uses metal nanoparticles (NPs) as catalytic seeds for CNT growth. However, these NPs remain as impurities in the grown CNT. In this study, nanodiamonds (NDs) with negligible catalytic activity were utilized as nonmetallic seeds instead of metal catalysts because they are stable at high temperatures and facilitate the growth of low-defect CNTs without residual metal impurities. Results demonstrate the successful growth of over 100-$μ$m-long CNTs by carefully controlling the growth conditions. Importantly, we developed an analysis method that utilizes secondary electron (SE) yield to distinguish whether or not CNTs grown from metal impurities. The absence of metallic NPs at the CNT tips was revealed by the SE yield mapping, whereas the presence of some kind of NPs at the same locations was confirmed by atomic force microscopy (AFM). These results suggest that most of the aligned CNTs were grown from nonmetallic seeds, most likely ND-derived NPs, via the tip-growth mode. Structural characterizations revealed the high crystallinity of CNTs, with relatively small diameters. This study presents the first successful use of nonmetallic seeds for kite growth and provides a convincing alternative for starting materials to prepare long, aligned CNTs without metal impurities. The findings of this study pave the way for more convenient fabrication of aligned CNT-based devices, potentially simplifying the production process by avoiding the need for the removal of metal impurities.

cond-mat.mtrl-sci↗

arXiv2023

Combination effect of growth enhancers and carbon sources on synthesis of single-walled carbon nanotubes from solid carbon growth seeds

In the synthesis of highly crystalline single-walled carbon nanotubes (SWCNTs), high growth temperatures are preferred, while the formation of impurity amorphous carbon (a-C) causes the termination of SWCNT growth and degrades its properties. To remove this by-product, H2O and CO2 have been employed in metal-catalyzed SWCNT growth systems because of their oxidizing ability. Recently, nonmetallic nanoparticles have become one of the growth seed candidates because of their high melting points and fewer metal impurities. In this study, by using nanodiamond-derived carbon nanoparticles as the solid growth seeds, we investigated the effect of CO2 and H2O on high-temperature SWCNT growth with two types of carbon-source supplies: C2H2 and C2H4. In this growth system, H2O showed oxidizing ability to etch a-C with either carbon sources. CO2 exhibited a similar a-C formation-preventing role in C2H4-supplied growth and achieved higher-purity SWCNTs with higher concentration of C2H4 than the case of H2O. However, in contrast to the other combinations, CO2 injection in the C2H2-supplied growth significantly enhanced the formation of a-C rather than the removal of it while the yield of SWCNTs was also increased, indicating the occurrence of the dehydration reaction between CO2 and C2H2. The present findings will lead to efficient growth of high-quality SWCNTs from nonmetallic growth seeds with the use of growth enhancers.

cond-mat.mtrl-sci↗