Tongxin Pan

Meituan LongCat Team, Bin Xiao, Chao Wang et al.

The prevailing Next-Token Prediction (NTP) paradigm has driven the success of large language models through discrete autoregressive modeling. However, contemporary multimodal systems remain language-centric, often treating non-linguistic modalities as external attachments, leading to fragmented architectures and suboptimal integration. To transcend this limitation, we introduce Discrete Native Autoregressive (DiNA), a unified framework that represents multimodal information within a shared discrete space, enabling a consistent and principled autoregressive modeling across modalities. A key innovation is the Discrete Native Any-resolution Visual Transformer (dNaViT), which performs tokenization and de-tokenization at arbitrary resolutions, transforming continuous visual signals into hierarchical discrete tokens. Building on this foundation, we develop LongCat-Next, a native multimodal model that processes text, vision, and audio under a single autoregressive objective with minimal modality-specific design. As an industrial-strength foundation model, it excels at seeing, painting, and talking within a single framework, achieving strong performance across a wide range of multimodal benchmarks. In particular, LongCat-Next addresses the long-standing performance ceiling of discrete vision modeling on understanding tasks and provides a unified approach to effectively reconcile the conflict between understanding and generation. As an attempt toward native multimodality, we open-source the LongCat-Next and its tokenizers, hoping to foster further research and development in the community. GitHub: https://github.com/meituan-longcat/LongCat-Next

Xinyu Zhou, Tongxin Pan, Lingyi Hong et al.

UAV tracking can be widely applied in scenarios such as disaster rescue, environmental monitoring, and logistics transportation. However, existing UAV tracking methods predominantly emphasize speed and lack exploration in semantic awareness, which hinders the search region from extracting accurate localization information from the template. The limitation results in suboptimal performance under typical UAV tracking challenges such as camera motion, fast motion, and low resolution, etc. To address this issue, we propose a dynamic semantic aware correlation modeling tracking framework. The core of our framework is a Dynamic Semantic Relevance Generator, which, in combination with the correlation map from the Transformer, explore semantic relevance. The approach enhances the search region's ability to extract important information from the template, improving accuracy and robustness under the aforementioned challenges. Additionally, to enhance the tracking speed, we design a pruning method for the proposed framework. Therefore, we present multiple model variants that achieve trade-offs between speed and accuracy, enabling flexible deployment according to the available computational resources. Experimental results validate the effectiveness of our method, achieving competitive performance on multiple UAV tracking datasets. The code is available at https://github.com/zxyyxzz/DSATrack.

Yanmeng Xing, Ye Sun, Tongxin Pan et al.

Establishing an independent academic identity is a central yet insufficiently understood challenge for early-career researchers. However, limited resources and mentor-driven research agendas often constrain early efforts toward autonomy. To provide large-scale quantitative evidence on how junior researchers develop independence, we introduce a framework that traces how mentees diverge from their mentors in both research topics and collaboration networks, and how these divergences relate to long-term scientific impact. Analyzing over 500,000 mentee-mentor pairs in Chemistry, Neuroscience, and Physics across six decades, we find that high-impact scientists often initiate work in secondary areas of their mentors' expertise while adaptively establishing distinct research trajectories. This pattern is most pronounced among mentees who eventually surpass their mentors' impact. We identify an inverted U-shaped relationship between topic divergence and mentees' enduring impact, with moderate divergence yielding the highest scientific impact, revealing an independence paradox in scientific careers. This pattern holds whether topic divergence is measured by citation network or semantic thematic distance. We further reveal that excessive direct mentor-mentee collaborations correlate with lower mentee impact, whereas expanding professional networks to include mentors' collaborators is beneficial. These findings not only offer actionable guidance for early-career researchers navigating independence but also inform institutional policies that promote mentorship structures supporting intellectual innovation and recognizing original contributions in promotion evaluations.