arXiv subjects

Sheng Jin

Publications and source records attributed to Sheng Jin.

At least 19 recordsLinked to original sources

arXiv2026

Thinking with Cameras: Active Visual Reasoning via Dynamic Viewpoint Control for Surveillance Video Understanding

Large vision-language models (LVLMs) have recently achieved remarkable progress in general-purpose video understanding. However, their application to surveillance videos remains challenging due to the lack of large-scale domain-specific datasets and the limitation of passive observation from fixed viewpoints. In surveillance scenarios, critical visual evidence can be easily missed when targets are distant, small, occluded, or move beyond the current camera view. In this work, we introduce CamVLM, a new framework for Thinking with Cameras, which enables LVLMs to actively acquire visual evidence through dynamic viewpoint control rather than passively analyzing fixed video streams. We first construct CCTV-Anomaly, a large-scale surveillance video understanding dataset containing 14,459 videos across 10 anomaly categories, with detailed captions and event annotations. We further formulate viewpoint control as an active visual perception problem and build CamTrack-53K, an object-centric viewpoint trajectory dataset for learning camera actions. Moreover, we propose a reinforcement learning based viewpoint policy optimization framework, which models camera control as a sequential decision-making process and learns long-horizon observation strategies beyond supervised trajectory imitation. Extensive experiments demonstrate that CamVLM achieves state-of-the-art performance under both passive observation and dynamic viewpoint settings, validating the effectiveness of active camera-based reasoning for surveillance video understanding. Our datasets, model, and code will be available at https://github.com/xiaozhang79/CamVLM .

cs.CV↗

arXiv2026

Prisma-World: Camera-Controllable Multi-Agent Video World Model

Video world models have made rapid progress in generating controllable visual experiences, but most of them still simulate the world from a single observer. Extending such models to multiple agents raises a central challenge: if each agent's future state is generated independently, overlapping views may instantiate different versions of the same scene, leading to inconsistent objects, layouts, and appearances across agents. Conventional camera conditioning controls individual trajectories, but it does not explicitly couple the generation of views that should agree under shared scene geometry. We introduce Prisma-World, a camera-controllable multi-agent world model that formulates multi-agent generation as a joint geometry-aware denoising process for cross-view consistency. Prisma-World processes all agent videos within one full-attention sequence, uses a multi-agent RoPE design to distinguish agent identities while preserving synchronized temporal coordinates, and injects relative camera geometry into attention to bias overlapping viewpoints toward shared scene evidence. To further strengthen multi-view consistency and enhance global spatial perception, we augment our framework with an overlap-decaying curriculum training paradigm alongside minimap-conditioned structural guidance. To facilitate the training and evaluation of multi-agent models, we introduce PrismaDataset, a large-scale UE5 dataset with panoramic acquisition across diverse scenes, composable multi-agent view groups with flexible agent counts and complex camera trajectories, and precise camera/action annotations for consistency training and evaluation. Experiments show that a single Prisma-World model can generate high-fidelity multi-agent videos with flexible agent numbers, camera controllability, improved cross-view consistency, and spatial grounding under minimap guidance.

cs.CV↗

arXiv2026

UniCAD: A Unified Benchmark and Universal Model for Multi-Modal Multi-Task CAD

Computer-Aided Design (CAD) underpins modern engineering and manufacturing by enabling the creation of precise, editable 3D models. However, CAD research typically studies tasks in isolation, and multi-modal, multi-task learning for CAD is hindered by the absence of a unified benchmark. To address this gap, we introduce UniCAD, a comprehensive benchmark for multi-modal CAD learning that covers point-to-CAD reconstruction, text/image-to-CAD generation, and CAD question answering across diverse input modalities. Alongside the benchmark, we present UniCAD-MLLM, a universal multi-modal large language model that ingests text, images, sketches, and point clouds and performs these heterogeneous tasks in an end-to-end fashion within a single framework. Extensive experiments on the UniCAD and Fusion360 benchmarks demonstrate that UniCAD-MLLM achieves state-of-the-art performance across all tasks, outperforming existing task-specific and multi-task baselines. We will release the dataset, code, and pretrained models to accelerate future research.

cs.CV↗

arXiv2026

Knowledge Index of Noah's Ark

Knowledge benchmarks for LLMs face three issues: scaling-driven designs that do not operationalize disciplinary representativeness; flat-payment annotation that permits lazy consensus; and unaudited ranking instability under bounded test budgets. We introduce KINA, an 899-item benchmark across 261 fine-grained disciplines, with two formal results. First, we cast representativeness as a coverage-style objective over expert-elicited anchors and operationalize disciplinary representativeness through a proxy, yielding a (1-1/e) greedy approximation (Proposition 1); the guarantee applies to the proxy, not to population representativeness. Second, we prove a bonus-on-bar tournament weakly FOSD-dominates flat payment in released-review quality, with incentive-compatibility threshold B > Delta C / Delta p_min (Theorem 1). Evaluating 42 models from 13 labs, the top model, Gemini-3.1-Pro-Preview, reaches 53.17%, followed by Claude-Opus-4.6 at 49.92% and GPT-5.4 at 48.55%, leaving substantial headroom below saturation. The full leaderboard shows a tiered structure rather than a smooth total order: a small frontier tier lies above 48%, a dense strong-model tier spans roughly 38-45%, and low-performing models remain only modestly above the 10% chance baseline. Tool augmentation adds up to 5.17 points across the five tool-use evaluations, with gains varying substantially across models. We report bootstrap ranking-stability statistics to make bounded-budget variance explicit and to discourage over-interpretation of adjacent ranks.

cs.AI↗

arXiv2026

Precision extraction of the deuteron electric polarizability via the Baldin sum rule with full low-energy coverage

The photodisintegration cross sections of the deuteron have been systematically measured over the photon energy range of 2.33-19.65 MeV at the Shanghai Laser Electron Gamma Source (SLEGS). By applying the well-established Baldin sum rule to the newly obtained data, the sum of the electric and magnetic dipole polarizabilities of the deuteron is extracted for the first time based solely on a dense and continuous experimental dataset, yielding {\alpha}E +\{beta}M = 0.719\pm0.009stat\pm0.014algo\pm0.023syst fm3 . With theoretical values of the magnetic polarizability \{beta}M calculated from the pionless effective field theory, a new value of the electric polarizability is obtained as {\alpha}E = 0.637 \pm 0.009stat \pm 0.014algo \pm 0.023syst \pm 0.004theo fm3 , which is in excellent agreement with current theoretical predictions. This result resolves the previous discrepancy between experimental measurements from elastic scattering and theory, providing a high-precision benchmark for nuclear interaction models.

nucl-ex↗

arXiv2026

Knowledge Visualization: A Benchmark and Method for Knowledge-Intensive Text-to-Image Generation

Recent text-to-image (T2I) models have demonstrated impressive capabilities in photorealistic synthesis and instruction following. However, their reliability in knowledge-intensive settings remains largely unexplored. Unlike natural image generation, knowledge visualization requires not only semantic alignment but also strict adherence to domain knowledge, structural constraints, and symbolic conventions, exposing a critical gap between visual plausibility and scientific correctness. To systematically study this problem, we introduce KVBench, a curriculum-grounded benchmark for evaluating knowledge-intensive T2I generation. KVBench covers six senior high-school subjects: Biology, Chemistry, Geography, History, Mathematics, and Physics. The benchmark consists of 1,800 expert-curated prompts derived from over 30 authoritative textbooks. Using this benchmark, we evaluate 14 state-of-the-art open- and closed-source models, revealing substantial deficiencies in logical reasoning, symbolic precision, and multilingual robustness, with open-source models consistently underperforming proprietary systems. To address these limitations, we further propose KE-Check, a two-stage framework that improves scientific fidelity via (1) Knowledge Elaboration for structured prompt enrichment, and (2) Checklist-Guided Refinement for explicit constraint enforcement through violation identification and constraint-guided editing. KE-Check effectively mitigates scientific hallucinations, narrowing the performance gap between open-source and leading closed-source models. Data and codes are publicly available at https://github.com/zhaoran66/KVBench.

cs.CV↗

arXiv2026

The Illusory Precision of TTV Masses: Hidden Solutions Behind Kepler-9's Tight Mass Ratio

Transit timing variations (TTV) are considered a tool for constraining the masses of transiting planets in the absence of radial-velocity data. Although theoretical studies have long revealed that TTV mass determinations intrinsically suffer from degeneracies, existing analyses of TTV data typically report a single-mode solution under a model with a specified number of planets. This is because fitting TTV curves in the high-dimensional solution space of TTV posterior is extremely challenging; even locating a single solution requires substantial computational resources. We developed an efficient mode-first searching algorithm that can locate multiple solutions in a single MCMC run. We applied this algorithm to Kepler-9 b and c, which have the highest-quality TTV data. We found that the observed TTV can be reproduced by many combinations of planetary masses spanning a broad range, rather than the previously assumed precise determination. The mass of Kepler-9 b can range from 31.6 to 47.1 $M_{\oplus}$, while that of Kepler-9 c can range from 21.8 to 32.3 $M_{\oplus}$, and even more broadly under looser constraints. These degenerate solutions follow a linear relationship under a tight mass ratio between the two planets, consistent with previous theoretical predictions. Furthermore, we demonstrate that achieving a globally converged posterior distribution for Kepler-9's TTV is impossible using a sampling algorithm that preserves the Markovian property. This underscores the need for caution when interpreting results from sampling algorithms that lack mathematical guarantees of global convergence.

astro-ph.EP↗

arXiv2026

Nii-body: Bayesian Inference of Multiplanet Dynamics via N-body Simulations

Many exoplanetary systems are multiplanet configurations whose long-term dynamics are governed by N-body gravitational interactions. Consequently, their detection signatures cannot be adequately described by Keplerian orbits. Accurately interpreting the observational data of these systems -- including radial velocity (RV), astrometry, and transit timing variations (TTVs) -- requires N-body integration. Motivated by this need, we developed a Bayesian fitting framework that couples N-body integration with Markov chain Monte Carlo (MCMC) to retrieve the system parameters of multiplanet systems. The code, named \texttt{Nii-body}, integrates an adaptive Runge--Kutta--Fehlberg 7(8) (RKF78) solver with an automated parallel tempering MCMC algorithm. Using simplified synthetic astrometric observations, we evaluated the efficiency and robustness of \texttt{Nii-body}'s N-body orbit retrieval on an idealized two-planet model, demonstrating its potential for future application to real observational data. The N-body fitting workflow can be readily extended to RV, TTVs, or combined datasets, providing a versatile engine for high-precision orbital inference in multiplanet systems.

astro-ph.EP↗

arXiv2026

Construction of MRD Codes Based on Circular-Shift Operations

Most well-known constructions of $(N \times n, q^{Nk}, d)$ maximum rank distance (MRD) codes rely on the arithmetic of $\mathbb{F}_{q^N}$, whose increasing complexity with larger $N$ hinders parameter selection and practical implementation. In this work, based on circular-shift operations, we present a construction of $(J \times n, q^{Jk}, d)$ MRD codes with efficient encoding, where $J$ equals to the Euler's totient function of a defined $L$ subject to $\gcd(q, L) = 1$. The proposed construction is performed entirely over $\mathbb{F}_q$ and avoids the arithmetic of $\mathbb{F}_{q^J}$. We further characterize the constructed MRD codes, Gabidulin codes and twisted Gabidulin codes using a set of $q$-linearized polynomials over the row vector space $\mathbb{F}_{q}^N$, and clarify their inherent difference and connection. For the case $J \neq m_L$, where $m_L$ denotes the multiplicative order of $q$ modulo $L$, we show that the proposed MRD codes, in a family of settings, are different from any Gabidulin code and any twisted Gabidulin code. For the case $J = m_L$, we prove that every constructed $(J \times n, q^{Jk}, d)$ MRD code coincides with a $(J \times n, q^{Jk}, d)$ Gabidulin code, yielding an equivalent circular-shift-based construction that operates directly over $\mathbb{F}_q$. In addition, we prove that under some parameter settings, the constructed MRD codes are equivalent to a generalization of Gabidulin codes obtained by summing and concatenating several $(m_L \times n, q^{m_Lk}, d)$ Gabidulin codes. When $q=2$, $L$ is prime and $n\leq m_L$, it is analyzed that generating a codeword of the proposed $((L-1) \times n, 2^{(L-1)k}, d)$ MRD codes requires $O(nkL)$ exclusive OR (XOR) operations, while generating a codeword of $((L-1) \times n, 2^{(L-1)k}, d)$ Gabidulin codes, based on customary construction, requires $O(nkL^2)$ XOR operations.

cs.IT↗

arXiv2026

Eccentricity Evolution of Warm Jupiters: The Role of Distant Perturbers and Nearby Companions

Warm Jupiters-giant exoplanets with orbital periods between 10 and 200 days-exhibit a broad range of eccentricities and are often accompanied by nearby low-mass planets. Understanding the origins of their orbital architectures requires examining both their migration histories and subsequent dynamical interactions. In this study, we perform extensive N-body simulations to explore how distant giant planet perturbers affect the eccentricity evolution of warm Jupiters and the role of nearby super-Earth companions in mediating these interactions. We find that while distant perturbers can induce large-amplitude eccentricity oscillations in warm Jupiters via the von Zeipel-Lidov-Kozai mechanism, the presence of nearby super-Earth companions often suppresses these variations via strong dynamical coupling. This mechanism naturally leads to a bimodal eccentricity distribution: warm Jupiters with nearby companions tend to maintain low eccentricities, whereas those without exhibit significantly broader eccentricity distributions. We show that reproducing the observed eccentricity distribution of warm Jupiters lacking nearby companions is most naturally explained if a substantial fraction of distant perturbers occupy dynamically extreme orbits, either with large mutual inclinations or high orbital eccentricities. These results support a scenario in which warm Jupiters experience substantial post-disk dynamical evolution, shaped jointly by distant perturbers and nearby companions.

astro-ph.EP↗

arXiv2026

Video-KTR: Reinforcing Video Reasoning via Key Token Attribution

Reinforcement learning (RL) has shown strong potential for enhancing reasoning in multimodal large language models, yet existing video reasoning methods often rely on coarse sequence-level rewards or single-factor token selection, neglecting fine-grained links among visual inputs, temporal dynamics, and linguistic outputs, limiting both accuracy and interpretability. We propose Video-KTR, a modality-aware policy shaping framework that performs selective, token-level RL by combining three attribution signals: (1) visual-aware tokens identified via counterfactual masking to reveal perceptual dependence; (2) temporal-aware tokens detected through frame shuffling to expose temporal sensitivity; and (3) high-entropy tokens signaling predictive uncertainty. By reinforcing only these key tokens, Video-KTR focuses learning on semantically informative, modality-sensitive content while filtering out low-value tokens. Across five challenging benchmarks, Video-KTR achieves state-of-the-art or highly competitive results, achieving 42.7\% on Video-Holmes (surpassing GPT-4o) with consistent gains on both reasoning and general video understanding tasks. Ablation studies verify the complementary roles of the attribution signals and the robustness of targeted token-level updates. Overall, Video-KTR improves accuracy and interpretability, offering a simple, drop-in extension to RL for complex video reasoning. Our code and models are available at https://github.com/zywang0104/Video-KTR.

cs.CV↗

arXiv2025

JavisGPT: A Unified Multi-modal LLM for Sounding-Video Comprehension and Generation

This paper presents JavisGPT, the first unified multimodal large language model (MLLM) for joint audio-video (JAV) comprehension and generation. JavisGPT has a concise encoder-LLM-decoder architecture, which has a SyncFusion module for spatio-temporal audio-video fusion and synchrony-aware learnable queries to bridge a pretrained JAV-DiT generator. This design enables temporally coherent video-audio understanding and generation from multimodal instructions. We design an effective three-stage training pipeline consisting of multimodal pretraining, audio-video fine-tuning, and large-scale instruction-tuning, to progressively build multimodal comprehension and generation from existing vision-language models. For instruction tuning, we construct JavisInst-Omni, a high-quality instruction dataset with over 200K GPT-4o-curated audio-video-text dialogues that cover diverse and multi-level comprehension and generation scenarios. On JAV comprehension and generation benchmarks, our experiments show that JavisGPT outperforms existing MLLMs, particularly in complex and temporally synchronized settings.

cs.CV↗

arXiv2025

EduGuardBench: A Holistic Benchmark for Evaluating the Pedagogical Fidelity and Adversarial Safety of LLMs as Simulated Teachers

Large Language Models for Simulating Professions (SP-LLMs), particularly as teachers, are pivotal for personalized education. However, ensuring their professional competence and ethical safety is a critical challenge, as existing benchmarks fail to measure role-playing fidelity or address the unique teaching harms inherent in educational scenarios. To address this, we propose EduGuardBench, a dual-component benchmark. It assesses professional fidelity using a Role-playing Fidelity Score (RFS) while diagnosing harms specific to the teaching profession. It also probes safety vulnerabilities using persona-based adversarial prompts targeting both general harms and, particularly, academic misconduct, evaluated with metrics including Attack Success Rate (ASR) and a three-tier Refusal Quality assessment. Our extensive experiments on 14 leading models reveal a stark polarization in performance. While reasoning-oriented models generally show superior fidelity, incompetence remains the dominant failure mode across most models. The adversarial tests uncovered a counterintuitive scaling paradox, where mid-sized models can be the most vulnerable, challenging monotonic safety assumptions. Critically, we identified a powerful Educational Transformation Effect: the safest models excel at converting harmful requests into teachable moments by providing ideal Educational Refusals. This capacity is strongly negatively correlated with ASR, revealing a new dimension of advanced AI safety. EduGuardBench thus provides a reproducible framework that moves beyond siloed knowledge tests toward a holistic assessment of professional, ethical, and pedagogical alignment, uncovering complex dynamics essential for deploying trustworthy AI in education. See https://github.com/YL1N/EduGuardBench for Materials.

cs.CL↗

arXiv2025

Beyond Correctness: Evaluating Subjective Writing Preferences Across Cultures

Current preference learning methods achieve high accuracy on standard benchmarks but exhibit significant performance degradation when objective quality signals are removed. We introduce WritingPreferenceBench, a dataset of 1,800 human-annotated preference pairs (1,200 English, 600 Chinese) across 8 creative writing genres, where responses are matched for objective correctness, factual accuracy, and length. On this benchmark, sequence-based reward models--the standard architecture for RLHF--achieve only 52.7% mean accuracy, while zero-shot language model judges perform at 53.9%. In contrast, generative reward models that produce explicit reasoning chains achieve 81.8% accuracy. We observe high within-model variance across genres: individual models range from 18.2% to 81.8% accuracy across different writing categories, with standard deviations averaging 10.1%. This variance persists regardless of model scale, with 27B parameter models showing no consistent improvement over 8B variants. Our results suggest that current RLHF methods primarily learn to detect objective errors rather than capture subjective quality preferences (e.g., creativity, stylistic flair, and emotional resonance), and that successful preference modeling may require intermediate reasoning representations rather than direct classification.

cs.CL↗

arXiv2025

COIG-Writer: A High-Quality Dataset for Chinese Creative Writing with Thought Processes

Large language models exhibit systematic deficiencies in creative writing, particularly in non-English contexts where training data is scarce and lacks process-level supervision. We present COIG-Writer, a novel Chinese creative writing dataset that captures both diverse outputs and their underlying thought processes through systematic reverse-engineering of high-quality texts. Unlike existing datasets that provide only input-output pairs, COIG-Writer comprises 1,665 meticulously curated triplets spanning 51 genres, each containing: (1) a reverse-engineered prompt, (2) detailed creative reasoning documenting decision-making processes, and (3) the final text. Through comprehensive experiments, we identify a two-component model of creative writing: narrative logic (provided by process supervision) and linguistic expression (maintained by general-purpose data). Our findings reveal three critical insights: (1) Process supervision is highly effective but requires stabilization with general data. A ratio of at least one creative sample to twelve general samples is needed to achieve optimal performance; below this threshold, the win rate progressively degrades (from 62.75% down to 35.78%)., (2) creative capabilities are culturally-bound with no cross-lingual transfer (89.26pp gap between Chinese and English performance), and (3) lexical diversity inversely correlates with creative quality (TTR paradox), suggesting high diversity signals compensatory behavior for logical deficiencies. These findings establish that creative excellence emerges from the interaction between logical scaffolding and linguistic grounding, analogous to how mathematical reasoning enhances but cannot replace linguistic competence in foundation models.

cs.CL↗

arXiv2025

Evolution in Simulation: AI-Agent School with Dual Memory for High-Fidelity Educational Dynamics

Large language models (LLMs) based Agents are increasingly pivotal in simulating and understanding complex human systems and interactions. We propose the AI-Agent School (AAS) system, built around a self-evolving mechanism that leverages agents for simulating complex educational dynamics. Addressing the fragmented issues in teaching process modeling and the limitations of agents performance in simulating diverse educational participants, AAS constructs the Zero-Exp strategy, employs a continuous "experience-reflection-optimization" cycle, grounded in a dual memory base comprising experience and knowledge bases and incorporating short-term and long-term memory components. Through this mechanism, agents autonomously evolve via situated interactions within diverse simulated school scenarios. This evolution enables agents to more accurately model the nuanced, multi-faceted teacher-student engagements and underlying learning processes found in physical schools. Experiment confirms that AAS can effectively simulate intricate educational dynamics and is effective in fostering advanced agent cognitive abilities, providing a foundational stepping stone from the "Era of Experience" to the "Era of Simulation" by generating high-fidelity behavioral and interaction data.

cs.AI↗

arXiv2025

High-Precision Measurement of D($\gamma$, $n$)$p$ Photodisintegration Reaction and Implications for Big-Bang Nucleosynthesis

We report on a high-precision measurement of the D($\gamma$,\,$n$)$p$ photodisintegration reaction at the newly commissioned Shanghai Laser Electron Gamma Source (SLEGS), employing a quasi-monochromatic $\gamma$-ray beam from Laser Compton Scattering. The cross sections were determined over $E_\gamma$=2.327--7.089 MeV, achieving up to a factor of 2.2 improvement in precision near the neutron separation threshold. Combined with previous data in a global Markov chain Monte Carlo (MCMC) analysis using dibaryon effective field theory, we obtained the unprecedentedly precise $p$($n$,\,$\gamma$)D cross sections and thermonuclear rate, with a precision up to $\approx$4 times higher than previous evaluations. Implemented in a standard Big-Bang Nucleosynthesis (BBN) framework, this new rate decreases uncertainty of the key cosmological parameter of baryon density $\Omega_b h^2$ by up to $\approx$16\% relative to the LUNA result. A residual $\approx$1.2$\sigma$ tension between $\Omega_b h^2$ constrained from primordial D/H observations and CMB measurements persists, highlighting the need for improved $dd$ reaction rates and offering potential hints of new physics beyond the standard model of cosmology.

nucl-ex↗

arXiv2025

EMNLP: Educator-role Moral and Normative Large Language Models Profiling

Simulating Professions (SP) enables Large Language Models (LLMs) to emulate professional roles. However, comprehensive psychological and ethical evaluation in these contexts remains lacking. This paper introduces EMNLP, an Educator-role Moral and Normative LLMs Profiling framework for personality profiling, moral development stage measurement, and ethical risk under soft prompt injection. EMNLP extends existing scales and constructs 88 teacher-specific moral dilemmas, enabling profession-oriented comparison with human teachers. A targeted soft prompt injection set evaluates compliance and vulnerability in teacher SP. Experiments on 14 LLMs show teacher-role LLMs exhibit more idealized and polarized personalities than human teachers, excel in abstract moral reasoning, but struggle with emotionally complex situations. Models with stronger reasoning are more vulnerable to harmful prompt injection, revealing a paradox between capability and safety. The model temperature and other hyperparameters have limited influence except in some risk behaviors. This paper presents the first benchmark to assess ethical and psychological alignment of teacher-role LLMs for educational AI. Resources are available at https://e-m-n-l-p.github.io/.

cs.CL↗