arXiv subjects

Hongkai Wang

Publications and source records attributed to Hongkai Wang.

5 recordsLinked to original sources

arXiv2026

MOSS-VL Technical Report

We present MOSS-VL, an open vision-language model family that treats real-time interaction -- perceiving while it speaks -- as a first-class capability. It is co-designed across the stack: the language decoder attends to vision only through gated cross-attention, so the model can naturally see incoming frames while generating; a synthesized interaction corpus supervises when to speak, when to stay silent, and when to revise; and a staged curriculum concentrates all real-time-specific training in one light final stage over a strong offline foundation. Offline, MOSS-VL-Instruct is competitive at comparable scale and leads temporal-reasoning video sets. Across four streaming benchmarks, MOSS-VL-Realtime posts the best average on three (second on the fourth) among open-source streaming models, sweeping the three subsets that squarely test proactive behavior -- 66.0 vs. 37.5 for the best baseline on OmniMMI Proactive Alerting. With 11.3B parameters but visual tokens outside the decoded sequence, MOSS-VL widens its time-to-first-token advantage over same-backbone Qwen3-VL-8B from 2.8x to 5.1x as visual context grows. We release all five checkpoints, the training curriculum, and the real-time inference code at https://github.com/OpenMOSS/MOSS-VL.

cs.CV↗

arXiv2026

MOSS-Video-Preview: Toward Real-Time Video Understanding via Cross-Attention

Video understanding is shifting from the offline paradigm -- taking a fully recorded video as input and producing a single answer after it ends -- toward real-time interaction, in which the model perceives new frames while still replying, revises its answer as new evidence appears, and remains silent when there is nothing to say. We present MOSS-Video-Preview to validate this paradigm. Our central claim is that perception must not be blocked by generation; its natural realization is a two-channel architecture. We argue that a cross-attention backbone is better suited to real-time vision-language fusion than the prevailing decoder-only design: visual features enter through a side channel rather than joining the autoregressive sequence, so perception and generation run on separate, non-blocking pathways -- reducing the frequency of visual processing and exposing a clean channel-wise interface for independent compression. We complement this with a data synthesis pipeline that converts dense captions into real-time understanding QA whose answers are revised to match what the model has perceived so far, and we specialize an offline model on these data to elicit real-time behavior. Our model trails the strong Qwen2.5-VL-7B baseline overall -- a gap we attribute primarily to data and scale rather than the architecture -- yet attains competitive offline video and multimodal understanding, remains robust on the spatial and fine-grained temporal reasoning central to real-time use, and acquires behaviors that offline models lack: continuous perception, answer revision, and timely silence. On a single H200 with 256 frames per video, it achieves about a 5x speedup in time to first token and 2.7x higher decoding throughput, with negligible degradation in offline ability. Our study of paradigm, architecture, and data outlines a viable path toward real-time video understanding.

cs.CV↗

arXiv2025

MAGI: Multi-Agent Guided Interview for Psychiatric Assessment

Automating structured clinical interviews could revolutionize mental healthcare accessibility, yet existing large language models (LLMs) approaches fail to align with psychiatric diagnostic protocols. We present MAGI, the first framework that transforms the gold-standard Mini International Neuropsychiatric Interview (MINI) into automatic computational workflows through coordinated multi-agent collaboration. MAGI dynamically navigates clinical logic via four specialized agents: 1) an interview tree guided navigation agent adhering to the MINI's branching structure, 2) an adaptive question agent blending diagnostic probing, explaining, and empathy, 3) a judgment agent validating whether the response from participants meet the node, and 4) a diagnosis Agent generating Psychometric Chain-of- Thought (PsyCoT) traces that explicitly map symptoms to clinical criteria. Experimental results on 1,002 real-world participants covering depression, generalized anxiety, social anxiety and suicide shows that MAGI advances LLM- assisted mental health assessment by combining clinical rigor, conversational adaptability, and explainable reasoning.

cs.CL↗

arXiv2017

Proton tracking in a high-granularity Digital Tracking Calorimeter for proton CT purposes

Radiation therapy with protons as of today utilizes information from x-ray CT in order to estimate the proton stopping power of the traversed tissue in a patient. The conversion from x-ray attenuation to proton stopping power in tissue introduces range uncertainties of the order of 2-3% of the range, uncertainties that are contributing to an increase of the necessary planning margins added to the target volume in a patient. Imaging methods and modalities, such as Dual Energy CT and proton CT, have come into consideration in the pursuit of obtaining an as good as possible estimate of the proton stopping power. In this study, a Digital Tracking Calorimeter is benchmarked for proof-of-concept for proton CT purposes. The Digital Tracking Calorimeteris applied for reconstruction of the tracks and energies of individual high energy protons. The presented prototype forms the basis for a proton CT system using a single technology for tracking and calorimetry. This advantage simplifies the setup and reduces the cost of a proton CT system assembly, and it is a unique feature of the Digital Tracking Calorimeter. Data from the AGORFIRM beamline at KVI-CART in Groningen in the Netherlands and Monte Carlo simulation results are used to in order to develop a tracking algorithm for the estimation of the residual ranges of a high number of concurrent proton tracks. The range of the individual protons can at present be estimated with a resolution of 4%. The readout system for this prototype is able to handle an effective proton frequency of 1 MHz by using 500 concurrent proton tracks in each readout frame, which is at the high end range of present similar prototypes. A future further optimized prototype will enable a high-speed and more accurate determination of the ranges of individual protons in a therapeutic beam.

physics.med-ph↗

arXiv2017

Comparison of machine learning methods for classifying mediastinal lymph node metastasis of non-small cell lung cancer from 18F-FDG PET/CT images

The present study shows that the performance of CNN is not significantly different from the best classical methods and human doctors for classifying mediastinal lymph node metastasis of NSCLC from PET/CT images. Because CNN does not need tumor segmentation or feature calculation, it is more convenient and more objective than the classical methods. However, CNN does not make use of the import diagnostic features, which have been proved more discriminative than the texture features for classifying small-sized lymph nodes. Therefore, incorporating the diagnostic features into CNN is a promising direction for future research.

cs.CV↗