arXiv subjects

Yutao Zhang

Publications and source records attributed to Yutao Zhang.

14 recordsLinked to original sources

arXiv2026

GLM-5V-Turbo: Toward a Native Foundation Model for Multimodal Agents

We present GLM-5V-Turbo, a step toward native foundation models for multimodal agents. As foundation models are increasingly deployed in real environments, agentic capability depends not only on language reasoning, but also on the ability to perceive, interpret, and act over heterogeneous contexts such as images, videos, webpages, documents, GUIs. GLM-5V-Turbo is built around this objective: multimodal perception is integrated as a core component of reasoning, planning, tool use, and execution, rather than as an auxiliary interface to a language model. This report summarizes the main improvements behind GLM-5V-Turbo across model design, multimodal training, reinforcement learning, toolchain expansion, and integration with agent frameworks. These developments lead to strong performance in multimodal coding, visual tool use, and framework-based agentic tasks, while preserving competitive text-only coding capability. More importantly, our development process offers practical insights for building multimodal agents, highlighting the central role of multimodal perception, hierarchical optimization, and reliable end-to-end verification.

cs.CV↗

arXiv2026

Collaborative Temporal Feature Generation via Critic-Free Reinforcement Learning for Cross-User Sensor-Based Activity Recognition

Human Activity Recognition using wearable inertial sensors is foundational to healthcare monitoring, fitness analytics, and context-aware computing, yet its deployment is hindered by cross-user variability arising from heterogeneous physiological traits, motor habits, and sensor placements. Existing domain generalization approaches either neglect temporal dependencies in sensor streams or depend on impractical target-domain annotations. We propose a different paradigm: modeling generalizable feature extraction as a collaborative sequential generation process governed by reinforcement learning. Our framework, CTFG (Collaborative Temporal Feature Generation), employs a Transformer-based autoregressive generator that incrementally constructs feature token sequences, each conditioned on prior context and the encoded sensor input. The generator is optimized via Group-Relative Policy Optimization, a critic-free algorithm that evaluates each generated sequence against a cohort of alternatives sampled from the same input, deriving advantages through intra-group normalization rather than learned value estimation. This design eliminates the distribution-dependent bias inherent in critic-based methods and provides self-calibrating optimization signals that remain stable across heterogeneous user distributions. A tri-objective reward comprising class discrimination, cross-user invariance, and temporal fidelity jointly shapes the feature space to separate activities, align user distributions, and preserve fine-grained temporal content. Evaluations on the DSADS and PAMAP2 benchmarks demonstrate state-of-the-art cross-user accuracy (88.53\% and 75.22\%), substantial reduction in inter-task training variance, accelerated convergence, and robust generalization under varying action-space dimensionalities.

cs.LG↗

arXiv2026

GLM-OCR Technical Report

GLM-OCR is an efficient 0.9B-parameter compact multimodal model designed for real-world document understanding. It combines a 0.4B-parameter CogViT visual encoder with a 0.5B-parameter GLM language decoder, achieving a strong balance between computational efficiency and recognition performance. To address the inefficiency of standard autoregressive decoding in deterministic OCR tasks, GLM-OCR introduces a Multi-Token Prediction (MTP) mechanism that predicts multiple tokens per step, significantly improving decoding throughput while keeping memory overhead low through shared parameters. At the system level, a two-stage pipeline is adopted: PP-DocLayout-V3 first performs layout analysis, followed by parallel region-level recognition. Extensive evaluations on public benchmarks and industrial scenarios show that GLM-OCR achieves competitive or state-of-the-art performance in document parsing, text and formula transcription, table structure recovery, and key information extraction. Its compact architecture and structured generation make it suitable for both resource-constrained edge deployment and large-scale production systems.

cs.CL↗

arXiv2026

The SAGES Critical View of Safety Challenge: A Global Benchmark for AI-Assisted Surgical Quality Assessment

Advances in artificial intelligence (AI) for surgical quality assessment promise to democratize access to expertise, with applications in training, guidance, and accreditation. This study presents the SAGES Critical View of Safety (CVS) Challenge, the first AI competition organized by a surgical society, using the CVS in laparoscopic cholecystectomy, a universally recommended yet inconsistently performed safety step, as an exemplar of surgical quality assessment. A global collaboration across 54 institutions in 24 countries engaged hundreds of clinicians and engineers to curate 1,000 videos annotated by 20 surgical experts according to a consensus-validated protocol. The challenge addressed key barriers to real-world deployment in surgery, including achieving high performance, capturing uncertainty in subjective assessment, and ensuring robustness to clinical variability. To enable this scale of effort, we developed EndoGlacier, a framework for managing large, heterogeneous surgical video and multi-annotator workflows. Thirteen international teams participated, achieving up to a 17% relative gain in assessment performance, over 80% reduction in calibration error, and a 17% relative improvement in robustness over the state-of-the-art. Analysis of results highlighted methodological trends linked to model performance, providing guidance for future research toward robust, clinically deployable AI for surgical quality assessment.

cs.CV↗

arXiv2026

GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning

We present GLM-4.1V-Thinking, GLM-4.5V, and GLM-4.6V, a family of vision-language models (VLMs) designed to advance general-purpose multimodal understanding and reasoning. In this report, we share our key findings in the development of the reasoning-centric training framework. We first develop a capable vision foundation model with significant potential through large-scale pre-training, which arguably sets the upper bound for the final performance. We then propose Reinforcement Learning with Curriculum Sampling (RLCS) to unlock the full potential of the model, leading to comprehensive capability enhancement across a diverse range of tasks, including STEM problem solving, video understanding, content recognition, coding, grounding, GUI-based agents, and long document interpretation. In a comprehensive evaluation across 42 public benchmarks, GLM-4.5V achieves state-of-the-art performance on nearly all tasks among open-source models of similar size, and demonstrates competitive or even superior results compared to closed-source models such as Gemini-2.5-Flash on challenging tasks including Coding and GUI Agents. Meanwhile, the smaller GLM-4.1V-9B-Thinking remains highly competitive-achieving superior results to the much larger Qwen2.5-VL-72B on 29 benchmarks. We open-source both GLM-4.1V-9B-Thinking and GLM-4.5V. We further introduce the GLM-4.6V series, open-source multimodal models with native tool use and a 128K context window. A brief overview is available at https://z.ai/blog/glm-4.6v. Code, models and more information are released at https://github.com/zai-org/GLM-V.

cs.CV↗

arXiv2025

OmniCache: A Trajectory-Oriented Global Perspective on Training-Free Cache Reuse for Diffusion Transformer Models

Diffusion models have emerged as a powerful paradigm for generative tasks such as image synthesis and video generation, with Transformer architectures further enhancing performance. However, the high computational cost of diffusion Transformers-stemming from a large number of sampling steps and complex per-step computations-presents significant challenges for real-time deployment. In this paper, we introduce OmniCache, a training-free acceleration method that exploits the global redundancy inherent in the denoising process. Unlike existing methods that determine caching strategies based on inter-step similarities and tend to prioritize reusing later sampling steps, our approach originates from the sampling perspective of DIT models. We systematically analyze the model's sampling trajectories and strategically distribute cache reuse across the entire sampling process. This global perspective enables more effective utilization of cached computations throughout the diffusion trajectory, rather than concentrating reuse within limited segments of the sampling procedure. In addition, during cache reuse, we dynamically estimate the corresponding noise and filter it out to reduce its impact on the sampling direction. Extensive experiments demonstrate that our approach accelerates the sampling process while maintaining competitive generative quality, offering a promising and practical solution for efficient deployment of diffusion-based generative models.

cs.CV↗

arXiv2025

Kimi-Audio Technical Report

We present Kimi-Audio, an open-source audio foundation model that excels in audio understanding, generation, and conversation. We detail the practices in building Kimi-Audio, including model architecture, data curation, training recipe, inference deployment, and evaluation. Specifically, we leverage a 12.5Hz audio tokenizer, design a novel LLM-based architecture with continuous features as input and discrete tokens as output, and develop a chunk-wise streaming detokenizer based on flow matching. We curate a pre-training dataset that consists of more than 13 million hours of audio data covering a wide range of modalities including speech, sound, and music, and build a pipeline to construct high-quality and diverse post-training data. Initialized from a pre-trained LLM, Kimi-Audio is continual pre-trained on both audio and text data with several carefully designed tasks, and then fine-tuned to support a diverse of audio-related tasks. Extensive evaluation shows that Kimi-Audio achieves state-of-the-art performance on a range of audio benchmarks including speech recognition, audio understanding, audio question answering, and speech conversation. We release the codes, model checkpoints, as well as the evaluation toolkits in https://github.com/MoonshotAI/Kimi-Audio.

eess.AS↗

arXiv2025

Muon is Scalable for LLM Training

Recently, the Muon optimizer based on matrix orthogonalization has demonstrated strong results in training small-scale language models, but the scalability to larger models has not been proven. We identify two crucial techniques for scaling up Muon: (1) adding weight decay and (2) carefully adjusting the per-parameter update scale. These techniques allow Muon to work out-of-the-box on large-scale training without the need of hyper-parameter tuning. Scaling law experiments indicate that Muon achieves $\sim\!2\times$ computational efficiency compared to AdamW with compute optimal training. Based on these improvements, we introduce Moonlight, a 3B/16B-parameter Mixture-of-Expert (MoE) model trained with 5.7T tokens using Muon. Our model improves the current Pareto frontier, achieving better performance with much fewer training FLOPs compared to prior models. We open-source our distributed Muon implementation that is memory optimal and communication efficient. We also release the pretrained, instruction-tuned, and intermediate checkpoints to support future research.

cs.LG↗

arXiv2025

An Innovative Brain-Computer Interface Interaction System Based on the Large Language Model

Recent advancements in large language models (LLMs) provide a more effective pathway for upgrading brain-computer interface (BCI) technology in terms of user interaction. The widespread adoption of BCIs in daily application scenarios is still limited by factors such as their single functionality, restricted paradigm design, weak multilingual support, and low levels of intelligence. In this paper, we propose an innovative BCI system that deeply integrates a steady-state visual evoked potential (SSVEP) speller with an LLM application programming interface (API). It allows natural language input through the SSVEP speller and dynamically calls large models to generate SSVEP paradigms. The command prompt, blinking frequency, and layout position are adjustable to meet the user's control requirements in various scenarios. More than ten languages are compatible with the multilingual support of LLM. A variety of task scenarios, such as home appliance control, robotic arm operation, and unmanned aerial vehicle (UAV) management are provided. The task interfaces of the system can be personalized according to the user's habits, usage scenarios, and equipment characteristics. By combining the SSVEP speller with an LLM, the system solves numerous challenges faced by current BCI systems and makes breakthroughs in functionality, intelligence, and multilingual support. The introduction of LLM not only enhances user experience but also expands the potential applications of BCI technology in real-world environments.

cs.HC↗

arXiv2024

A Survey of Neural Network Robustness Assessment in Image Recognition

In recent years, there has been significant attention given to the robustness assessment of neural networks. Robustness plays a critical role in ensuring reliable operation of artificial intelligence (AI) systems in complex and uncertain environments. Deep learning's robustness problem is particularly significant, highlighted by the discovery of adversarial attacks on image classification models. Researchers have dedicated efforts to evaluate robustness in diverse perturbation conditions for image recognition tasks. Robustness assessment encompasses two main techniques: robustness verification/ certification for deliberate adversarial attacks and robustness testing for random data corruptions. In this survey, we present a detailed examination of both adversarial robustness (AR) and corruption robustness (CR) in neural network assessment. Analyzing current research papers and standards, we provide an extensive overview of robustness assessment in image recognition. Three essential aspects are analyzed: concepts, metrics, and assessment methods. We investigate the perturbation metrics and range representations used to measure the degree of perturbations on images, as well as the robustness metrics specifically for the robustness conditions of classification models. The strengths and limitations of the existing methods are also discussed, and some potential directions for future research are provided.

cs.CV↗

arXiv2024

The particle acceleration study in blazar jet

The particle acceleration of blazar jets is crucial to high-energy astrophysics, yet the acceleration mechanism division in blazar subclasses and the underlying nature of these mechanisms remain elusive. In this work, we utilized the synchrotron spectral information (synchrotron peak frequency, $\log ν_{\rm sy}$, and corresponding curvature, $b_{\rm sy}$) of 2705 blazars from the literature and studied the subject of particle acceleration in blazar jets by analysing the correlation between $\log ν_{\rm sy}$ and $1/b_{\rm sy}$. Our results suggested that the entire sample follows an energy-dependent probability acceleration (EDPA). Specifically, the low inverse Compton peak sources (LCPs) follow the mechanism that fluctuations of fractional gain acceleration (FFGA), while the high inverse Compton peak sources (HCPs) follow an acceleration mechanism of EDPA. Our results indicated that the separation between LCPs and HCPs results from the electron peak Lorentz factor ($γ_{\rm p}$), and the differentiation should originate from different acceleration mechanisms. Moreover, our study revealed a transition in the acceleration mechanism from FFGA to EDPA around $\log ν_{\rm sy} \sim 15$ through a detailed analysis of binned-$\log ν_{\rm sy}$. The mechanism of FFGA dominates the particle acceleration in LCP jets because of stronger jets and the EDPA dominates the particle energy gain in the HCPs due to a more efficient acceleration process.

astro-ph.HE↗

arXiv2020

Comparison between $Fermi$ Detected and non-$Fermi$ Detected Superluminal Sources

Active galactic nuclei (AGNs) have been attracting research attention due to their special observable properties. Specifically, a majority of AGNs are detected by Fermi-LAT missions, but not by Fermi-LAT, which raises the question of whether any differences exist between the two. To answer this issue, we compile a sample of 291 superluminal AGNs (189 FDSs and 102 non-FDSs) from available multi-wavelength radio, optical, and X-ray (or even $γ$-ray) data and Doppler factors and proper motion ($μ$) (or apparent velocity ($β_{\rm{app}}$)); calculated the apparent velocity from their proper motion, Lorentz factor ($Γ$), viewing angle ($ϕ$) and co-moving viewing angle ($ϕ_{co}$) for the sources with available Doppler factor ($δ$); and performed some statistical analyses for both types. Our study indicated that1. In terms of average values, FDSs have higher proper motions ($μ$), apparent velocities ($β_{\rm app}$), Doppler factor ($δ$), Lorentz factor ($Γ$), and smaller viewing angle ($ϕ$). Nevertheless, there is no clear difference in co-moving viewing angles ($ϕ_{\rm co}$).

astro-ph.HE↗

arXiv2020

A framework for constructing a huge name disambiguation dataset: algorithms, visualization and human collaboration

We present a manually-labeled Author Name Disambiguation(AND) Dataset called WhoisWho, which consists of 399,255 documents and 45,187 distinct authors with 421 ambiguous author names. To label such a great amount of AND data of high accuracy, we propose a novel annotation framework where the human and computer collaborate efficiently and precisely. Within the framework, we also propose an inductive disambiguation model to classify whether two documents belong to the same author. We evaluate the proposed method and other state-of-the-art disambiguation methods on WhoisWho. The experiment results show that: (1) Our model outperforms other disambiguation algorithms on this challenging benchmark. (2) The AND problem still remains largely unsolved and requires more in-depth research. We believe that such a large-scale benchmark would bring great value for the author name disambiguation task. We also conduct several experiments to prove our annotation framework could assist annotators to make accurate results efficiently and eliminate wrong label problems made by human annotators effectively.

cs.SI↗

arXiv2014

Social Network Integration: Towards Constructing the Social Graph

In this work, we formulate the problem of social network integration. It takes multiple observed social networks as input and returns an integrated global social graph where each node corresponds to a real person. The key challenge for social network integration is to discover the correspondences or interlinks across different social networks. We engaged an in-depth analysis across three online social networks, AMiner, Linkedin, and Videolectures in order to address what reveals users' social identity, whether the social factors consistent across different social networks and how we can leverage these information to perform integration. We proposed a unified framework for the social network integration task. It crawls data from multiple social networks and further discovers accounts correspond to the same real person from the obtained networks. We use a probabilistic model to determine such correspondence, it incorporates features like the consistency of social status and social ties across different, as well as one-to-one mapping constraint and logical transitivity to jointly make the prediction. Empirical experiments verify the effectiveness of our method.

cs.SI↗