Searcharxiv⌕ Search

arXiv subjects

Zekun Li

Publications and source records attributed to Zekun Li.

At least 19 recordsLinked to original sources

CAM3DNet: Comprehensively mining the multi-scale features for 3D Object Detection with Multi-View Cameras

Query-based 3D object detection methods using multi-view images often struggle to efficiently leverage dynamic multi-scale information, e.g., the relationship between the object features and the geometric of the queries are not sufficiently learned, directly exploring the multi-scale spatiotemporal features will pay too many costs. To address these challenges, we propose CAM3DNet, a novel sparse query-based framework which combines three new modules, composite query (CQ), adaptive self-attention (ASA), and multi-scale hybrid sampling (MSHS). First, the core idea in the CQ module is a multi-scale projection strategy to transform 2D queries into 3D space. Second, the ASA module learns the interactions between the spatiotemporal multi-scale queries. Third, the MSHS module uses the deformable attention mechanism to sample multi-scale object information by considering multi-scales queries, pyramid feature maps, and 2D-camera prior knowledge. The entire model employs a backbone network and a feature pyramid network (FPN) as the encoder, then introduces a YOLOX and a DepthNet as a ROI\_Head to produce CQ, and repeatedly utilizes ASA and MSHS as the decoder to gain detection features. Extensive experiments on the nuScenes, Waymo, and Argoverse benchmark datasets demonstrate the effectiveness of our CAM3DNet, and most existing camera-based 3D object detection methods are outperformed. Besides, we make comprehensive ablation studies to check the individual effect of CQ, ASA, and MSHS, as well as their cost of space and computation complexity.

cs.CV↗

TrainExplorer: Autonomous LLM Post-Training via Plan-and-Decouple Exploration

LLM-based agents can automate model post-training, accelerating progress toward recursive self-improvement. Despite their promise, existing approaches remain under-explored and face two main challenges: (1) strategy lock-in that limits exploration beyond the initial strategy; and (2) coupled execution that obscures the effects of individual modifications. To address these challanges, we propose TrainExplorer, a multi-agent framework for autonomous post-training, that incorporates strategy planning and execution decoupling, to achieve broaden strategy exploration. First, strategy planning draws on scientific literature to construct a candidate search space of training strategies. Then, decoupled execution enables single-component interventions and attribution to guide strategy exploration. Extensive experiments on PostTrainBench demonstrate that TrainExplorer achieves a weighted aggregate score of 39.02, outperforming Claude Code using the same LLM by 7.32 and surpassing GPT-5.6 Sol-powered Codex, which scores 36.23. Code will be available.

cs.LG↗

EQUIP-RPT: A framework for objective evaluation of quantitative imaging methods in radiopharmaceutical therapies

Quantitative imaging (QI) provides a non-invasive approach for dosimetry and quantitative response assessment in radiopharmaceutical therapies (RPTs). QI methods are being developed for instrumentation, acquisition, reconstruction, and image analysis. However, a key barrier to routine adoption of these methods is that evidence demonstrating reliable performance of QI methods on clinically relevant quantification tasks in RPTs remains limited. Our goal is to outline the Evaluation of Quantitative Imaging Performance in Radiopharmaceutical Therapy (EQUIP-RPT) framework, a structured approach for objective, task-based evaluation of QI methods for RPTs. We outline the EQUIP-RPT framework based on image-science and task-based image-quality evaluation methodology, practical considerations, and recent literature. The framework is organized around four linked design steps: (i) specification of the quantification task, (ii) design of the study population and imaging protocol, (iii) task-specific information extraction, and (iv) selection of figures of merit (FoMs). The EQUIP-RPT framework is presented in the context of four quantification tasks in RPT: activity uptake, absorbed dose, volumetric tumor burden, and lesion-level longitudinal quantifications. The framework distinguishes two paradigms for task-specific information extraction and organizes FoMs according to the performance attributes they characterize. We describe key elements to apply the EQUIP-RPT framework in simulation, physical phantom, and clinical studies. Finally, we show how EQUIP-RPT can be used to generate a quantitative claim based on the evaluation study. The EQUIP-RPT framework is designed to provide a structured approach for objective task-based evaluation of QI methods in RPTs, thereby facilitating clinical translation of these methods.

physics.med-ph↗

Prospects for probing supermassive binary black holes in strongly lensed active galactic nuclei via microlensing effects

Gravitational microlensing in strongly lensed active galactic nuclei (AGNs) provides a powerful probe of accretion-disk structure and may offer a unique opportunity to identify sub-parsec supermassive binary black holes (SMBBHs) through distinctive signatures. Upcoming wide-field surveys are expected to discover a large sample of strongly lensed AGNs, a fraction of which may host SMBBHs detectable via microlensing. In this paper, we employ semi-analytic models to assess the detection prospects of sub-parsec SMBBHs in lensed AGNs, focusing on systems with resolvable multiple images suitable for microlensing analysis. Accounting for galaxy-scale strong-lensing statistics, and including the cluster-lensing contribution as an optimistic statistical extension, we estimate that up to approximately 87 such systems could be spatially resolved by individual upcoming surveys, including the Vera C. Rubin Observatory (Rubin), Euclid, China's Space Survey Telescope (CSST), and the Roman. For the Rubin WFD survey, we estimate that periodic oscillations in microlensing light curves could identify $\sim0.12-0.68$ short-period ($<5$\,yr) binaries, whereas diagnostics based on the reconstructed size-wavelength relation apply to a broader population, potentially identifying $\sim0.4-5.3$ systems, depending on the model assumptions. CSST, Euclid, and Roman may also reveal up to $10.6$, $8.9$, and $1.2$ systems exhibiting size-wavelength features, and $2.3$, $1.9$, $0.3$ systems showing periodic oscillations, which should be regarded as imaging-selected candidates whose microlensing-based identification requires adequate time-domain follow-up, including cadence, monitoring baseline, photometric precision, and reliable time-delay measurements. Our results show that strongly lensed AGNs provide a promising, though observationally demanding, pathway for finding candidate SMBBHs via microlensing.

astro-ph.GA↗

MM-VeriAgent: Learning to Use Extensive Tools to Verify Multimodal Misinformation with Reinforcement Learning

Real-world multimodal misinformation often involves mixed forgery sources, requiring sample-specific detection strategies. Existing tool-augmented methods rely on predefined workflows or inference-time planning, limiting adaptability or increasing inference cost. To address this issue, we introduce \textbf{MM-VeriAgent}, which learns to verify mixed-source multimodal misinformation with tools. We first build \textbf{MM-VeriTools}, a specialized toolkit for misinformation detection agents. By benchmarking various candidate models and methods on the sub-tasks required by mixed-source detection, we select the strongest for textual, visual, and cross-modal forgery analysis and encapsulate them as callable tools with a unified interface. On top of this toolkit, we train the LVLM agent with reinforcement learning to teach it how to use these tools to better solve mixed-source detection. Since many of the tools are specialized models whose online execution at every rollout severely limits RL efficiency, we further introduce \textbf{Tool-Execution Cache}, which pre-executes candidate tool calls and reuses their cached outputs during training. This preserves multi-step rollouts while reducing online tool execution, largely improving the training efficiency.Experiments on MMFakeBench demonstrate substantial accuracy gains over the base model without explicit tool search at inference time. Ablation and efficiency analyses further validate the learned tool-use policy and show that Tool-Execution Cache reduces online tool executions during training.

cs.CV↗

ScenePercept: Scenario-Conditioned Perceptual Assessment of Real-World User-Generated Content Images

The rapid development of mobile imaging and social platforms has led to explosive growth in user-generated content images, making their systematic and comprehensive perceptual assessment increasingly important. Existing image perception benchmarks are typically task-oriented, curating specific scenarios for quality or aesthetic assessment and applying a uniform perceptual criterion across diverse scenarios. This limits their ability to capture the scene diversity and perceptual complexity of real-world UGC images. To address these limitations, we introduce ScenePercept-Bench, a benchmark designed to reflect the distribution of real-world UGC images, comprising 23,100 images across 21 categories and 105 fine-grained scenes. To account for the influence of scene context on human perception, we design a hierarchical, scene-aware annotation protocol, covering visual quality, aesthetic perception, and scene semantics, with 15 attributes in total. Each image is annotated with an overall score, 15 attribute-level scores, and structured explanatory reasoning. Building on this benchmark, we further introduce Scenario-Conditioned Perceptual Alignment, a training framework that tackles the heterogeneity of perceptual criteria across scenarios. SCPA jointly optimizes absolute score prediction, preference ranking, and cross-scenario calibration, preserving the model general perceptual assessment ability while encouraging it to explore and adapt perceptual criteria across different scenarios. Extensive experiments demonstrate that SCPA achieves state-of-the-art performance in UGC image perception and generalizes effectively to established IQA and IAA benchmarks.

cs.CV↗

World-T3: Test-Time Training as Persistent Memory for Real-Time Video World Models

Recent advances in video world models have enabled plausible interactive environments, yet long-horizon generation places a fundamental demand on visual memory: the previously observed world must remain consistent as the model continues to generate. Existing autoregressive approaches often implicitly rely on the generative model itself to retain such information, leading to spatial memory decay beyond the active context. We introduce World-T3, a camera-controlled video world model with a test-time training memory that consolidates past observations into dedicated model parameters. To robustly update this memory, we propose a recursive least-squares rule that incrementally absorbs novel visual features via online pose-keyed memory updates. Our TTT memory reads out previously observed scene features from the fast-weight state by camera pose, enabling consistent generation with constant storage and per-step computation regardless of rollout length. To facilitate real-time interactive generation with this robust memory, we further distill the autoregressive model into a few-step generator via dual-teacher distribution matching. On a long-horizon revisit benchmark, World-T3 raises revisit PSNR from 13.52 to 18.31 dB and reduces absolute rotation error from 3.344 to 1.075 relative to the strongest baseline on both metrics, while sustaining 20.6 FPS at 1280 x 704 on a single NVIDIA B200 GPU.

cs.CV↗

Sonic-AMG: Streaming Human Motion Generation from Spatial Audio

Modeling human reactions to spatial audio requires understanding both what is heard and where it comes from. Prior work generates motion offline from complete acoustic and spatial input sequences. Humans, however, react to sounds without knowing what they will hear next. We present Sonic-AMG, to our knowledge the first framework for streaming human motion generation from spatial audio. The model encodes acoustic content and body-relative source geometry as separate tokens. Conditioned on these tokens, a causal transformer with a flow head generates continuous motion latents, from which a causal decoder reconstructs motion segments. Standard teacher-forced training derives motion histories and body-relative source locations from ground-truth motion, whereas streaming inference relies on generated motion for both, resulting in a training-inference mismatch. To reduce this history discrepancy, we introduce rehearsal training, which combines full-sequence teacher-forced supervision with supervised prediction along short autoregressive self-rollouts. To keep spatial conditioning consistent with generated motion, we propose spatial feedback, which recomputes body-relative source locations from generated root poses during both rehearsal training and inference. On SAM, Sonic-AMG achieves state-of-the-art FID and R-precision. It further supports real-time online generation of coherent reactions across sound events and source locations. Project page: https://jingchengni.com/sonic-amg/.

cs.CV↗

Probing anomalous $Wtb$ couplings through $bW$ spin correlations in top-quark pair production at the HL-LHC

We propose a spin-density-matrix framework for probing anomalous $Wtb$ interactions in top-quark pair production by exploiting the polarizations of the $b$ quark and the $W$ boson and their spin correlations. The $b$-quark spin is analyzed through angular distributions in semileptonic $Λ_b$ decay following fragmentation, while the leptonic $W$ decay serves as a $W$-polarization analyzer. We derive the $bW$ spin density matrix in the presence of anomalous $Wtb$ couplings and show that its polarization and spin-correlation parameters provide linear sensitivity to these couplings. We construct experimentally accessible angular observables associated with these parameters and evaluate their statistical sensitivities. For the HL-LHC at $\sqrt{s}=14~ \mathrm{TeV}$ with an integrated luminosity of $3000~ \mathrm{fb}^{-1}$, the proposed method yields improved constraints on most anomalous $W t b$ couplings, with the sensitivity to $V_R$ enhanced by a factor of 8 and that to $g_L$ by a factor of about 6.

hep-ph↗

Can LLMs Predict the Future? A Brier Score Analysis of Prediction Markets

We study whether model upgrades improve probability estimates for prediction-market questions. Our Resolved Market Forecasting (RMF) benchmark contains 3,000 resolved binary questions across nine domains, on which we evaluate six Claude and Qwen model variants using a question-only, zero-shot protocol. We assess Brier scores relative to an empirical base-rate predictor, examine their Murphy decomposition, and compare models through paired differences, with results stratified by event category and timing relative to training cutoffs. In the reported post-cutoff stratum, the four Claude models achieve Brier scores of 0.183-0.192, improving on the base-rate reference by 0.024-0.033. Qwen 32B does not significantly outperform that reference, although its paired Brier is 0.024 lower than that of the 7B checkpoint. The evaluated Claude version and tier upgrades yield no significant improvement. Within-model differences across event categories exceed the observed differences among Claude variants. These results show why forecasting scores should be interpreted alongside simple probability baselines and question composition: under this protocol, newer versions or higher model tiers do not consistently produce more accurate probabilities.

cs.AI↗

Task-Oriented Key-Layer KV Communication for Efficient Latent Multi-Agent Collaboration

Large language model-based multi-agent systems improve complex problem solving through collaboration, while latent communication directly transmits model internal states to avoid the high inference costs of natural language. However, existing KV-based latent communication methods prioritize sender-side state fidelity, leading to substantial communication and computation overhead and potentially introducing redundant information. To address these limitations, we revisit latent communication from a task-oriented perspective, shifting its objective from sender-side state fidelity to receiver-side task sufficiency. Under this formulation, we propose KITE, a training-free framework for task-oriented key-layer KV communication. KITE identifies a task-effective key layer using a receiver trajectory distortion criterion, transmits only the latent working memory associated with the key layer, and further uses the same layer as the entry point for autoregressive latent reasoning. Experiments on seven benchmarks across two model families and three model scales show that, compared with full-layer KV communication, KITE reduces communication volume by 28-36$\times$, achieves up to 3$\times$ end-to-end inference speedup, and improves accuracy by up to 23.3 percentage points.

cs.LG↗

MoSAT: Human Motion Generation from Spatial Audio and Textual Description

Human motion is shaped by both external acoustic events and behavioral intent: spatial audio conveys environmental cues that elicit or guide a response, while text specifies the desired action and how it should be performed. In this paper, we study the novel task of human motion synthesis jointly conditioned on spatial audio and natural language, a problem that has been largely overlooked in previous research. To support this task, We introduce STAM, a dataset of motion sequences paired with spatial audio and detailed textual annotations whose rich vocabulary affords precise and nuanced specification of human motions. We further introduce MoSAT, a latent flow-matching framework for full-body motion generation jointly conditioned on natural-language intent and directional spatial-audio cues through hierarchical cross-attention before generating motion. Such a hierarchical design enhances temporally coherent and semantically aligned motion sequences. We also develop tri-modal evaluators for comprehensive evaluation on this novel task. Extensive experiments show that MoSAT achieves the SOTA performance by leveraging spatial audio's intrinsic motion-shaping properties alongside textual semantics, enabling precise and diverse motion in various scenarios.

cs.GR↗

A Population Study for Searching Supermassive Binary Black Holes in Active Galactic Nuclei: Continuum Spectral Features and Periodic Variabilities

Active sub-pc supermassive binary black holes (SMBBHs) are expected to exhibit various electromagnetic signatures due to unique dynamical and geometric structures of their accretion, but observational identification of them remains challenging. In this paper, we adopt semianalytic models to investigate both deficits in spectral-energy-distributions (SEDs) of these systems induced by gaps/holes in their accretion disks and periodic variations in their light curves induced by either orbital-modulated Doppler boosting or accretion rate variation of each SMBH component. We construct a population model to generate SMBBHs across cosmic time by considering their orbital evolution and associated accretion and radiation processes. By estimating the continuum emission from each mock system and its variation, we investigate the detection of SMBBHs via either SED-deficit signature or light curve periodicity under reasonably given criteria. We find that all-sky surveys with filters similar to those of the China Space Station Telescope or Rubin/LSST could identify up to approximately $3\times 10^2$ SMBBHs via SED-deficit features and/or $10^3$ SMBBHs via periodicity ($\lesssim5$\,yr), although both selections may suffer from a high rate of false positives. A few to a dozen SED-deficit SMBBHs may be detected by future pulsar timing arrays with signal-to-noise ratio $\gtrsim3$, enabling multi-messenger observations. Only $\sim20\%-53\%$ of SED-deficit selected SMBBHs may also be detected via periodic variations, and $\sim7\%-9\%$ of periodic variation selected SMBBHs may be detected via SED-deficit signatures. The false positives for those SMBBHs selected jointly by both methods are negligible, highlighting the importance of searching for active SMBBH systems using joint methods.

astro-ph.HE↗

Flavor--Kinetic Entanglement Production from Decay and Scattering at Finite Density

We extend the scattering-entanglement dictionary to finite-density environments by investigating the flavor--kinetic bipartition of the Hilbert space. We show that tracing over kinematic degrees of freedom maps the total branch-changing transition probability directly onto the leading flavor--kinetic linear entanglement entropy. At finite density, the vacuum branch-changing probability is replaced by an occupation-weighted collision probability, built from the same directed reaction-density kernel that enters the integrated Boltzmann equation. The resulting observable is the bath-averaged flavor--kinetic entanglement entropy of a pair sampled from the medium. As a proof of principle, this framework is applied to an $O(N)$ singlet-scalar extended model to probe thermal phase transitions. In the examples studied, the resulting entanglement entropy serves as a collision-based phase-transition-type diagnostic, exhibiting a finite discontinuity across a first-order phase transition and a nonanalytic temperature derivative for continuous transitions. These examples suggest a novel way to characterize thermal phase structures, distinct from traditional thermodynamic order parameters.

hep-ph↗

HumanCLAW: Can Vision-Language Models Act Through a Body?

Evaluating whether a vision-language model (VLM) can act through a physical body is challenging. The outcome of an action couples the VLM's decision with motor control. When a task fails, it is hard to tell whether the VLM made a bad choice or the motor controller simply failed to execute it, e.g., losing balance and falling. In this work, we introduce HumanCLAW, an evaluation framework that decouples action decision-making from low-level execution. At every step, a harnessed, off-the-shelf VLM issues an atomic skill command, and the command is translated into a sub-second chunk of continuous full-body motion with real physical consequences, including gravity and collisions. The body can therefore act freely in the physical world, while execution-side disturbances, balance and motor errors, are factored out. What remains measurable is the model's action intelligence: its moment-to-moment choice of what the body should execute next. Based on this framework, we build HumanCLAW-Bench: 1,218 long-horizon, egocentric find-navigate-interact episodes across 41 indoor scenes. We test nine state-of-the-art VLMs and find that none solves the benchmark; the best model reaches only a 16.8% success rate. Recognizing the target is not the bottleneck. What current VLMs lack is embodied self-awareness: they lose track of their own body, failing to tell where it is, whether it has reached the goal, or whether it has hit an obstacle.

cs.CV↗

Collider Spin Tomography with Missing Neutrinos

Missing neutrinos need not destroy collider spin tomography. We formulate the visible measurement under kinematic ambiguities arising from invisible particles as a coarse-grained positive-operator-valued measure on the production spin density matrix. We show that information loss is governed by the null space of the resulting visible-data map, not by the number of kinematic solutions. In $e^+e^-\toτ^+τ^-\toπ^+π^-+ν\barν$, the twofold ambiguity leaves only the antisymmetric spin-correlation combination $C_{nr}-C_{rn}$ unidentifiable, while the differential production rate and the remaining fourteen spin coefficients are identifiable. For practical reconstruction under kinematic ambiguities, we develop a self-consistent fixed-point unfolding method using only visible data, without assuming a theoretical production template. Closure tests in Standard Model and anomalous tau-dipole benchmarks show that the method reproduces the truth-level differential production rate and all identifiable spin coefficients, whereas the usual flat average over kinematic folds gives significantly biased reconstructions. When a nontrivial null space is present, the reconstructed identifiable subspace together with positivity yields controlled ranges for concurrence and the CHSH parameter.

hep-ph↗

Ms. Forcing: Efficient Streaming Video Generation with Multi-Scale Patchification and Attention

Streaming video diffusion models have made substantial progress toward interactive and dynamic world simulation, but the nested autoregressive and denoising loops of conventional next-frame generation hinder real-time deployment. Recent rolling-window methods pipeline denoising across multiple consecutive frames at different noise levels, improving throughput and long-horizon stability. However, they tokenize every state at the same fine spatial granularity, leaving substantial noise-dependent redundancy in the joint denoising window. We propose Ms.Forcing, an efficient streaming video generation paradigm that adapts spatial granularity to each state's noise level. Its Multi-Scale Patchification (MSP) assigns coarser patches to noisier states, reducing the active-window token count by 45%, while Multi-Scale Self-Attention (MSSA) matches the density of visible non-sink keys and values to each query scale to further reduce attention cost. Because both schedules are fixed by window position, Ms.Forcing retains a static, hardware-friendly computation graph. We further introduce Homogeneous-Noise-Level DMD (H-DMD), which assembles each fake video from clean predictions sharing the same source noise level, thereby reducing the mismatch between DMD training sequences and inference-time rollouts. The multi-scale design helps offset the additional training cost of backpropagating through overlapping windows. We include both quantitative and qualitative experiments to show that Ms.Forcing reaches 22.84 FPS on a single H200 GPU, 39.6% faster than Rolling Forcing, while significantly improving VBench scores in both short video and long video generation setting.

cs.CV↗

Deform360: A Massive Multi-view Visuotactile Dataset for Deformable World Models

Predicting object dynamics (i.e., world modeling) is a fundamental challenge for robotic manipulation, and modeling deformable objects presents a particularly difficult case due to their high-dimensional state spaces and complex material properties. While current world models approach this through two distinct paradigms: learning the dynamics over the 2D pixel space or more explicit 3D geometric space. A systematic understanding of their relative strengths and limitations remains elusive due to the lack of diverse, large-scale real-world data. To address this, we present Deform360, a large-scale visuotactile dataset featuring 198 daily-life objects, 1,980 interaction sequences, and over 215 hours of observations from 41 surround-view cameras and bimanual tactile grippers to capture both global motion and contact-induced local deformations. Leveraging a novel markerless visuotactile 3D tracking pipeline to extract dense geometry and motion, we systematically evaluate current state-of-the-art world models, comparing 2D video models against 3D particle models. Finally, we provide a preliminary demonstration indicating the real-world applicability of our dataset by performing robot planning tasks on deformable objects. Our analysis reveals key insights into the trade-offs between structural priors and scalability, providing a solid benchmark for future research in generalizable deformable object-centric world modeling. Project website: https://deform360.lhy.xyz

cs.RO↗