Magnetic particle imaging (MPI) is an innovative imaging modality offering high spatio-temporal resolution for reconstructing magnetic particle distributions. To achieve high-quality MPI images, traditional methods such as the X-space method and system matrix (SM) method operate in the time and frequency domains, respectively. Both time and frequency domain signals essentially consist of harmonic components, meaning that the quality of the reconstruction is closely tied to the accuracy of these harmonic elements. However, the presence of background noise and the complexities associated with SM collection significantly hinder harmonic quality, leading to degradation of reconstruction fidelity. To tackle these challenges, we propose a unified framework based on harmonic knowledge to enhance the quality of reconstruction. In particular, our approach involves pretraining autoencoders by restoring masked SMs, thereby modeling the relationship between harmonics. Subsequently, using appropriate decoders, the pretrained encoder can be transferred to tasks such as spectrum denoising and SM super-resolution. Our framework’s effectiveness is validated on these two tasks through simulation and publicly available datasets, where it consistently outperforms state-of-the-art (SOTA) methods. In addition, our framework shows marked improvements in reconstruction quality for both time and frequency domain data collected with our in-house MPI system.

We introduce Kwai Keye-VL-2.0-30B-A3B, an open-source Mixture-of-Experts (MoE) multimodal foundation model designed to advance long-video understanding and agentic intelligence. To address the challenges of ultra-long contexts, information redundancy, and prohibitive computational costs inherent in hour-level videos, Keye-VL-2.0 is the first to adapt DeepSeek Sparse Attention (DSA) to GQA-based multimodal architectures, enabling lossless 256K context processing while capturing critical frames and long-range temporal dependencies. This architecture is underpinned by a highly optimized training and inference infrastructure, including scalable video I/O, heterogeneous ViT-LM parallelism, and custom DSA kernels that significantly maximize throughput and minimize computational overhead. Furthermore, to overcome the algorithmic dilemma of catastrophic forgetting during multi-task alignment, we introduce Cross-Modal Multi-Teacher On-Policy Distillation (MOPD) paired with Context-RL and Video-RL. By distilling dense token-level teacher feedback from on-policy rollouts back into the MoE backbone, which activates only 3B parameters, Keye-VL-2.0 natively empowers advanced agent collaboration across Code, Tool, and Search scenarios with multimodal self-correction. Extensive evaluations across video understanding, temporal grounding, reasoning, STEM, and agent benchmarks demonstrate that Keye-VL-2.0-30B-A3B achieves state-of-the-art performance among models of similar scale, particularly excelling in fine-grained temporal localization on TimeLens and long-video comprehension on Video-MME-v2 and LongVideoBench. We release our model checkpoints to accelerate community progress toward scalable and robust multimodal agentic applications.

We present KAT-Coder-V2, an agentic coding model developed by the KwaiKAT team at Kuaishou. KAT-Coder-V2 adopts a "Specialize-then-Unify" paradigm that decomposes agentic coding into five expert domains - SWE, WebCoding, Terminal, WebSearch, and General - each undergoing independent supervised fine-tuning and reinforcement learning, before being consolidated into a single model via on-policy distillation. We develop KwaiEnv, a modular infrastructure sustaining tens of thousands of concurrent sandbox instances, and scale RL training along task complexity, intent alignment, and scaffold generalization. We further propose MCLA for stabilizing MoE RL training and Tree Training for eliminating redundant computation over tree-structured trajectories with up to 6.2x speedup. KAT-Coder-V2 achieves 79.6

Magnetars are the most strongly magnetized compact objects known in the Universe and are regarded as one of the primary engines powering a variety of enigmatic, high-energy transients. However, our understanding of magnetars remains highly limited, constrained by observational sample size and radiative variability. XTE J1810–197, which re-entered a radio-active phase in 2018, is one of only six known radio-pulsating magnetars. Leveraging the distinctive capability for simultaneous dual-frequency observations, we utilized the Shanghai Tianma Radio Telescope (TMRT) to monitor this magnetar continuously at both 2.25 and 8.60 GHz, capturing its entire evolution from radio activation to quenching. This enabled precise characterization of the evolution in its integrated profile, spin frequency, flux density, and spectral index (α, defined by S ∝ fα). The first time derivative of its spin frequency v̇ passed through four distinct phases—rapid decrease, violent oscillation, steady decline, and stable recovery—before returning to its pre-outburst value concomitant with the cessation of radio emission. Remarkably, both the amplitudes and the characteristic time-scales of these v̇ variations match those observed during the previous outburst that began in 2003, providing the first demonstration that post-outburst rotational evolution and radiative behavior in a magnetar are repeatable. A twisted-magnetosphere model can qualitatively account for this repeatability as well as for the progressive narrowing and abrupt disappearance of the radio pulse radiation, thereby receiving strong observational support.

Ground-based time-domain observatories require minute-by-minute, site-scale awareness of cloud cover, yet existing all-sky datasets are short, daylight-biased, or lack astrometric calibration. We present LenghuSky-8, an eight-year (2018-2025) all-sky imaging dataset from a premier astronomical site, comprising 429,620 512 × 512 frames with 81.2

Generative recommendation models in the OneRec family have been widely deployed in many real-world services, such as short-video, live-streaming, advertising, and e-commerce. However, these generative models can only benefit from the scaling advantage, while their reasoning ability is hard to activate, since we cannot construct meaningful Chain-of-Thought (CoT) sequences consisting of itemic tokens only. Inspired by the success of the reasoning-style “think before answer” paradigm in the LLM field, we conduct preliminary studies (i.e., OneRec-Think, OpenOneRec) to explore reasoning capability in generative recommendation. Nevertheless, we notice an unexpected phenomenon: the thinking mode does not show advantages over the non-thinking mode. Drawing insights from recent findings on CoT robustness in multi-modal language models, we argue that effective reasoning in recommendation rests on two factors: perception, the ability to ground itemic tokens in their underlying language semantics, and cognition, the ability to reorganize a user's behavior sequence into coherent latent interest points. We therefore propose OneReason, which includes: (1) strong itemic token perception in pre-training, (2) a three-level cognition-enhanced CoT format for recommendation tasks in SFT, and (3) a specialize-then-unify training recipe in RL to enhance the thinking ability.

Site characterization studies have identified Lenghu site as a world-class location for optical-infrared astronomy owing to exceptional atmospheric conditions. The spatial distribution of precipitable water vapor (PWV) and its altitudinal variations are crucial for observatory siting strategies. Utilizing multi-year meteorological data from five high-elevation weather stations (3311-4200 m), this study quantifies key spatiotemporal variations in PWV. Our analysis indicates a clear trend of decreasing PWV with increasing elevation across all times of day, with the frequency of sub-2 mm events at night rising from 44.59% at 3311 m to 53.63% at 4200 m, suggesting improved suitability of these sites for infrared-to-submillimeter observations. Pronounced seasonality manifests as consistent summer maxima and winter minima across altitudes, with the 4200 m site exhibiting notably arid winters averaging 0.91 mm contrasting with summer means of 7.03 mm. These findings establish an essential reference for altitudinal optimization of next-generation astronomical facilities at Lenghu.

Mixture-of-Experts (MoE) scales language models efficiently through sparse expert activation, and its dynamic variant further reduces computation by adjusting the activated experts in an input-dependent manner. Existing dynamic MoE methods usually rely on pre-training from scratch or task-specific adaptation, leaving the practical conversion of fully trained MoE underexplored. Enabling such adaptation would directly alleviate the inference costs by allowing easy tokens to bypass unnecessary expert during serving. This paper introduces Zero-Expert Self-Distillation Adaptation (ZEDA), a low-cost framework that transforms post-trained static MoE models into efficient dynamic ones. To stabilize this architectural conversion, ZEDA injects parameter-free zero-output experts into each MoE layer and adapts the augmented model through two-stage self-distillation, utilizing the original MoE as a frozen teacher and applying a group-level balancing loss. On Qwen3-30B-A3B and GLM-4.7-Flash across 11 benchmarks spanning math, code, and instruction following, ZEDA eliminates over 50% of expert FLOPs at marginal accuracy loss. It outperforms the strongest dynamic MoE baseline by 6.1 and 4.0 points on the two models, and delivers ~1.20$\times$ end-to-end inference speedup.

Addressing the challenges of fragmented task definitions and the heterogeneity of unstructured data in multimodal parsing, this paper proposes the Omni Parsing framework. This framework establishes a Unified Taxonomy covering documents, images, and audio-visual streams, introducing a progressive parsing paradigm that bridges perception and cognition. Specifically, the framework integrates three hierarchical levels: 1) Holistic Detection, which achieves precise spatial-temporal grounding of objects or events to establish a geometric baseline for perception; 2) Fine-grained Recognition, which performs symbolization (e.g., OCR/ASR) and attribute extraction on localized objects to complete structured entity parsing; and 3) Multi-level Interpreting, which constructs a reasoning chain from local semantics to global logic. A pivotal advantage of this framework is its evidence anchoring mechanism, which enforces a strict alignment between high-level semantic descriptions and low-level facts. This enables “evidence-based” logical induction, transforming unstructured signals into standardized knowledge that is locatable, enumerable, and traceable. Building on this foundation, we constructed a standardized dataset and released the Logics-Parsing-Omni model, which successfully converts complex audio-visual signals into machine-readable structured knowledge. Experiments demonstrate that fine-grained perception and high-level cognition are synergistic, effectively enhancing model reliability. Furthermore, to quantitatively evaluate these capabilities, we introduce OmniParsingBench. Code, models and the benchmark are released at https://github.com/alibaba/Logics-Parsing/tree/master/Logics-Parsing-Omni.

We present Qwen-Image-2.0, an omni-capable image generation foundation model that unifies high-fidelity generation and precise image editing within a single framework. Despite recent progress, existing models still struggle with ultra-long text rendering, multilingual typography, high-resolution photorealism, robust instruction following, and efficient deployment, especially in text-rich and compositionally complex scenarios. Qwen-Image-2.0 addresses these challenges by coupling Qwen3-VL as the condition encoder with a Multimodal Diffusion Transformer for joint condition-target modeling, supported by large-scale data curation and a customized multi-stage training pipeline. This enables strong multimodal understanding while preserving flexible generation and editing capabilities. The model supports instructions of up to 1K tokens for generating text-rich content such as slides, posters, infographics, and comics, while significantly improving multilingual text fidelity and typography. It also enhances photorealistic generation with richer details, more realistic textures, and coherent lighting, and follows complex prompts more reliably across diverse styles. Extensive human evaluations show that Qwen-Image-2.0 substantially outperforms previous Qwen-Image models in both generation and editing, marking a step toward more general, reliable, and practical image generation foundation models.

Scale-up fabrics connecting GPUs and AI accelerators carry tensor transfers together with remote reads, writes, atomics, and notifications over shared target-side receiver resources. Byte-denominated credits protect link buffers and streaming HBM traffic, but poorly represent small operations dominated by Atomic execution or response injection. This paper presents SemaCredit, a receiver controller that admits each remote-memory operation against a vector of target-resource demands and returns each component when its corresponding HBM, Atomic, or response stage completes. In a deterministic event simulator with multipath queues, eight HBM partitions, a serialized Atomic engine, and a response engine, SemaCredit matches a strong per-resource byte baseline on HBM-hotspot traffic while reducing small-operation P99 latency by 52.4

Scale-up accelerator fabrics send latency-sensitive flits over serial links at hundreds of gigabits per second. Their reliability pipeline first relies on FEC, then detects residual failures and replays unacknowledged data. At these line rates, delayed feedback lets later flits enter the replay window before a residual failure is reported, so standard replay can amplify one corrupted flit into a suffix retransmission. This paper presents PREFACE, a pre-FEC controller for temporally correlated burst errors. A two-state Bayesian filter converts corrected-symbol observations into a next-flit burst posterior and jointly selects FEC strength with an outstanding-flit cap. We implement PREFACE in ns-3 with publicly verifiable UALink 200G 1.0 replay semantics. PREFACE improves goodput by 10.52

Large Vision-Language Models (LVLMs) demonstrate strong performance on high-level semantic tasks but lack fine-grained perception for pixel-level understanding, particularly in Referring Expression Segmentation (RES). The core challenge is representing irregular object contours compatibly with the sequential nature of language models. We introduce Seg-LLaVA, an end-to-end LVLM that redefines segmentation by predicting contour points in polar coordinates rather than dense pixel masks. Our Polar Coordinate Adaptive Sampling (PCAS) samples key boundary points, providing a unified representation that enhances training stability and shape fidelity. A lightweight refinement module leverages hidden states to generate high-precision masks with improved boundaries. We also introduce LAGS, a large-scale dataset enabling complex, interactive, and multi-object video segmentation. Extensive experiments show Seg-LLaVA achieves state-of-the-art performance, substantially surpassing previous methods in segmentation accuracy, localization precision, and language grounding.

Although diffusion models have approached humanlevel performance in text-to-speech (TTS) tasks, their iterative inference process-often requiring hundreds of steps-remains a bottleneck for real-time deployment. VoiceFlow addresses this issue by leveraging flow matching and flow rectification to compress sampling into just 2-10 steps, enabling high-quality speech synthesis at low step counts. However, this method requires frame-level duration annotations prior to training and relies on deterministic duration prediction, limiting its ability to model natural rhythm variations and reducing flexibility. To overcome these limitations, we propose VoiceFlow++, which introduces a stochastic duration predictor, the Monotonic Alignment Search algorithm, and adversarial training to achieve expressive, highfidelity, and ultra-fast speech synthesis. Experiments on the LJSpeech and LibriTTS datasets demonstrate that VoiceFlow++ can generate natural and expressive mel-spectrograms with minimal inference steps, significantly outperforming existing fast TTS methods. Our approach provides a scalable and efficient framework for real-time deployment in interactive systems.

Residential energy analytics has advanced with increasing availability of energy data, e.g., from smart meters, to provide new insights into consumption behaviors and enhance energy efficiency and management. This review synthesizes public datasets, machine learning methods, and applications across two major analytical approaches: behavioral analysis through clustering and load disaggregation using Non-Intrusive Load Monitoring (NILM). We examine how characteristics of residential energy data are used in these two analytical approaches and analyze how data properties, including temporal resolution, population scale, and measurement granularity, relate to reported applications and performance. A structured analysis of 24 major residential energy datasets reveals distinct data utilization patterns across reported applications. Studies demonstrating consumption behavioral analysis through clustering typically utilize population scale datasets with medium temporal resolution for customer segmentation and demand reduction. In contrast, studies achieving effective load disaggregation through NILM often rely on high-frequency measurements for appliance-level disaggregation. These patterns reflect prevailing data collection and usage practices, while both analytical approaches continue to expand through integration of hybrid models and privacy-preserving training. This review integrates methodological development, energy applications, and dataset utilization, providing a comprehensive understanding of residential energy analytics and future research directions. Continued progress is needed to address challenges related to data scarcity, methodological adaptability, and deployment constraints, including advances in privacy-preserving frameworks, such that residential energy efficiency and management can be further enhanced across diverse residential contexts.

The integration of neural networks into actor-critic frameworks has been pivotal in advancing the field of reinforcement learning, enabling agents to perform complex tasks with greater efficiency and adaptability. However, neural network-based actor-critic models remain opaque ``black boxes,'' concealing their decision-making processes and hindering their use in critical applications where transparent and explainable reasoning is essential. This work introduces an innovative adaptation of the actor-critic framework that unites neural networks with rule ensembles to tackle key challenges in reinforcement learning. We harness the computational power, scalability, and adaptability of neural networks to model the critic, while integrating a rule ensemble system for the actor, ensuring transparency and interpretability for decision-making. Our study establishes a theoretical foundation for integrating rule ensembles into the Advantage Actor-Critic (A2C) framework. Experimental results from seven classic and complex environments demonstrate that our proposed method matches or exceeds the performance of representative RL models, including symbolic methods, while offering self-interpretability and transparency.

The design of XPU-PIM integrated Domain-Specific Accelerators relies heavily on high-performance simulators. However, existing simulation tools often suffer from limited integration capability and insufficient simulation speed. To address these challenges, this paper proposes HOPESim, a lightweight and fast simulator specifically designed for XPU-PIM accelerators. We introduce a Register Synchronous Simulation Model (RSSM), which abstracts multiple Processing Elements (PEs) and their interconnections into parallel blocks, enabling seamless integration with third-party simulators through a modular structure. A non-global timestamp strategy is proposed to break cycle dependencies between components, while a decoupled lightweight coroutine library leveraging modern C++ features is adopted to enhance large-scale simulation performance. Experimental results demonstrate that HOPESim achieves over 27× speedup compared to RTL simulation while maintaining compatibility with third-party simulation tools.

TimeGAN, an advanced framework for time-series generation, remains insufficiently examined from a privacy perspective. This paper systematically evaluates its vulnerability to membership inference attacks (MIAs) and introduces four attack strategies under different knowledge settings. In the white-box scenario, we design a discriminator-based mean scoring method and an embedding-based method leveraging temporal representations via an LSTM classifier, both achieving over 92% accuracy and AUCs up to 98.64% on benchmark datasets. In the black-box scenario, we propose a shadow discriminator approach and a similarity-based attack using Dynamic Time Warping (DTW), with DTW substantially outperforming Euclidean distance and other baselines, reaching 75.60% accuracy. These results highlight heterogeneous privacy leakage channels in TimeGAN and underscore the importance of robust defenses for time-series generative modeling.

A world model predicts environment dynamics based on current observations and actions, serving as a core cognitive mechanism for reasoning and planning. In this work, we investigate how world modeling based on language models can further push the boundaries of general agents. (i) We first focus on building foundation models for agentic environment simulation. We introduce Qwen-AgentWorld-35B-A3B and Qwen-AgentWorld-397B-A17B, the first language world models capable of simulating agentic environments covering 7 domains via long chain-of-thought reasoning. Leveraging more than 10M environment interaction trajectories of 7 domains in real-world environments, we develop Qwen-AgentWorld through a three-stage training pipeline: CPT injects general-purpose world modeling capabilities from the state transition dynamics and augmented professional corpora, SFT activates next-state-prediction reasoning, and RL sharpens simulation fidelity through a tailored framework with hybrid rubric-and-rule rewards. To evaluate language world models, we present AgentWorldBench, a comprehensive benchmark constructed from real-world interactions of 5 frontier models on 9 established benchmarks. Empirical results demonstrate that Qwen-AgentWorld significantly outperforms existing frontier models. (ii) Beyond foundation models, we further investigate two complementary paradigms through which world modeling enhances general agents. First, as a decoupled environment simulator, Qwen-AgentWorld supports scalable and controllable simulation of thousands of real-world environments for agentic RL, yielding gains that surpass real-environment training alone. Second, as a unified agent foundation model, world-model training acts as a highly effective warm-up that improves downstream performance across 7 agentic benchmarks. Code: https://github.com/QwenLM/Qwen-AgentWorld

The lower mantle is the layer with the largest volume and mass proportion among Earth's internal layers. Studies on the detection, distribution patterns, and heterogeneity of lower mantle heterogeneities will help reveal the mantle composition, mineral phase transitions, and thermochemical structure of Earth ' s interior, thereby deepening our understanding of geodynamic processes such as mantle rheology and mantle convection. Northeast China is located at the subduction front of the northwestern Pacific Plate and has undergone multiple large-scale tectonic events since the Mesozoic, including the subduction of the Paleo-Pacific Plate (Izanagi Plate) and the Pacific Plate. The studies of deep structures beneath Northeast China indicate that subducted slabs have stagnated in the mantle transition zone or penetrated into the lower mantle. Due to method limitations, the detection of small-scale lower mantle scatterers has mostly been confined to regions with intermediate- and deep-focus earthquakes in subduction zones. In this study, we extend the receiver function analysis to detect the lower mantle scatterers, and successfully find the scatterers beneath the continental region of Northeast China. Based on dense broadband seismic arrays deployed on Northeast China, we extract P-wave receiver functions and cross-validated the reliability of lower mantle scatterer' signals previously detected in intermediate- and deep-focus earthquake zones through slowness analysis, 3D velocity correction, and frequency-dependent analysis. Additionally, we newly identified multiple lower mantle scatterers at depths of 900 similar to 1000 km beneath the intracontinental region of Northeast China. Combined tomography models, we interpret these scatterers as remnants of the Izanagi Plate subduction, suggesting that the influence of the Izanagi Plate subduction extends at least 1000 km west and north of the Sea of Japan. Our study extends the traditional receiver function analysis to the deeper lower mantle, confirming its effectiveness in detecting lower mantle scatterers. It provides novel, robust, and independent seismic evidence for observations of scatterers in the uppermost lower mantle. Furthermore, it proposes a new approach for tracking remnants of ancient subducted slabs in the deep mantle, offering fresh insights into the dynamics of plate subduction processes.