arXiv subjects

Yidan Zhang

Publications and source records attributed to Yidan Zhang.

At least 19 recordsLinked to original sources

arXiv2026

RSMeM: Knowledge-Enhanced Memory Evolution for Remote Sensing Agents with Systematic Evaluation

Geoscience research requires complex analysis and domain expertise, with remote sensing (RS) observations as a key foundation. However, existing RS agents built on general-purpose LLMs remain largely domain-agnostic, resulting in brittle and error-prone workflows. Moreover, these failures are seldom consolidated into a reusable experience for subsequent analyses. To address this issue, we introduce RSMeM, a knowledge-enhanced memory evolution mechanism that bootstraps RS agents with pre-distilled domain knowledge and iteratively integrates online experience for robust multi-step tool execution. RSMeM is composed of two components: (i) Hierarchical Knowledge Grounding, which performs taxonomy-aware retrieval over a hierarchical domain corpus to guide planning and tool selection; and (ii) Failure-Aware Experience Refinement, which distills failure-annotated tool-use traces into reusable constraints for next-round tool execution. By iteratively employing these two processes, RS agents can evolve to absorb task-level domain knowledge and effectively translate it into instance-level execution experience. Extensive experiments on EarthBench demonstrate that RSMeM consistently improves tool-use performance and end-to-end answer across a diverse set of LLM backbones. Notably, RSMeM achieves a 6% accuracy improvement on DeepSeek-V3.2 with less than 1% additional experience tokens, demonstrating the strong knowledge density of our distilled experience.

cs.AI↗

arXiv2026

GeoForge: Non-Parametric Self-Evolving Agents for Earth-Observation Reasoning

Earth observation (EO) agents construct scientifically valid tool workflows and ground their conclusions in current geospatial evidence. This is challenging because EO workflows are constrained by sensing semantics, product dependencies, spatial and temporal compatibility, and parameter requirements. Existing agents often search a broad operation space for each query, while recent self-evolving systems do not fully organize heterogeneous EO trajectories into reusable knowledge across different decision levels. To solve this problem, we present GeoForge, a training-free, self-evolving framework that transforms completed trajectories into a structured nonparametric execution state. GeoForge constrains the operation space according to the sensing context, then retrieves a task-conditioned prior from three complementary memories. Workflow Graph Memory captures global operation order, Action-Level Experiences provide local corrections, and the Adapted Skill Standard Operating Procedure preserves procedural and data constraints. The retrieved prior guides tool execution, while current observations remain the basis of the final answer. After each task, a safety-gated distillation process converts grounded trajectories into reusable execution knowledge for future retrieval. This execution, distillation, and reuse loop improves planning without updating the backbone LLM. Experiments on multiple geospatial benchmarks demonstrate that GeoForge consistently improves both task accuracy and tool-use trajectory quality across diverse LLM backbones, while substantially reducing tool-planning and reasoning errors for most LLMs.

cs.AI↗

arXiv2026

On the large-time asymptotics of the defocusing mKdV equation with step-like initial data

We study the Cauchy problem for the defocusing modified Korteweg-de Vries (mKdV) equation with step-like initial data approaching nonzero constants $c_l$ and $c_r$ as $x \to -\infty$ and $x\to+\infty$, respectively. Assuming $c_l>c_r>0$, the solution exhibits a rarefaction wave structure. We first develop the inverse scattering transform for the solution satisfying these step-like boundary conditions. Using the associated scattering data, we prove that there exists a unique global solution of the Cauchy problem and characterize it in terms of a Riemann-Hilbert (RH) problem. By applying the nonlinear steepest descent method to this RH problem, we rigorously obtain large-time asymptotics of rarefaction wave solution in three distinct space-time regions, each characterized by a different leading order behavior. They are: (I) a left-field region where the solution approaches the left background constant, modulo a small oscillatory correction, (II) a central region where the solution exhibits a slowly varying profile that transitions between the two constants, and (III) a right-field region where the solution tends to the right background constant, up to an algebraically small correction. Rigorous derivations of the leading terms, sub-leading terms as well as the error bounds are presented.

math.AP↗

arXiv2026

Painlevé \uppercase\expandafter{\romannumeral34\relax} and collisionless shock in the defocusing NLS equation with step-like initial data in the transition regions

We consider the Cauchy problem for the defocusing nonlinear Schrödinger (NLS) equation with step-like initial data. Using the nonlinear steepest descent method, we derive the long-time asymptotic expansion of the solution to the Cauchy problem in three distinct transition regions. In the first two transition regions, the leading-order asymptotics are characterized by Painlevé \uppercase\expandafter{\romannumeral34\relax}-type formula, while in the third one is a collisionless shock region, the leading-order asymptotics is describedin terms of Riemann theta functions. Our analysis is based on the Riemann-Hilbert formulation associated with the Cauchy problem of the defocusing NLS equation.

math.AP↗

arXiv2026

Painleve XXXIV asymptotics for the defocusing mKdV equation with step-like initial data in transition regions

In this paper, we consider the Cauchy problem for the defocusing modified Korteweg-de Vries (mKdV) equation with a step-like initial data. Based on the Riemann-Hilbert problem associated with the mKdV equation, we derive the long-time asymptotic expansion of the solution to the defocusing mKdV equation in two transition regions using the nonlinear steepest descent method. It comes out that the leading term in the expansion is shown to match the corresponding background constants. The subleading term, however, decays at the order $\mathcal{O}(t^{-2/3})$, and its coefficient is derived from the associated Painlevé \uppercase\expandafter{\romannumeral34\relax} model.

math.AP↗

arXiv2026

Generative World Renderer

Scaling generative inverse and forward rendering to real-world scenarios is bottlenecked by the limited realism and temporal coherence of existing synthetic datasets. To bridge this persistent domain gap, we introduce a large-scale, dynamic dataset curated from visually complex AAA games. Using a novel dual-screen stitched capture method, we extracted 4M continuous frames (720p/30 FPS) of synchronized RGB and five G-buffer channels across diverse scenes, visual effects, and environments, including adverse weather and motion-blur variants. This dataset uniquely advances bidirectional rendering: enabling robust in-the-wild geometry and material decomposition, and facilitating high-fidelity G-buffer-guided video generation. Furthermore, to evaluate the real-world performance of inverse rendering without ground truth, we propose a novel VLM-based assessment protocol measuring semantic, spatial, and temporal consistency. Experiments demonstrate that inverse renderers fine-tuned on our data achieve superior cross-dataset generalization and controllable generation, while our VLM evaluation strongly correlates with human judgment. Combined with our toolkit, our forward renderer enables users to edit styles of AAA games from G-buffers using text prompts.

cs.CV↗

arXiv2026

Cheers: Decoupling Patch Details from Semantic Representations Enables Unified Multimodal Comprehension and Generation

A recent cutting-edge topic in multimodal modeling is to unify visual comprehension and generation within a single model. However, the two tasks demand mismatched decoding regimes and visual representations, making it non-trivial to jointly optimize within a shared feature space. In this work, we present Cheers, a unified multimodal model that decouples patch-level details from semantic representations, thereby stabilizing semantics for multimodal understanding and improving fidelity for image generation via gated detail residuals. Cheers includes three key components: (i) a unified vision tokenizer that encodes and compresses image latent states into semantic tokens for efficient LLM conditioning, (ii) an LLM-based Transformer that unifies autoregressive decoding for text generation and diffusion decoding for image generation, and (iii) a cascaded flow matching head that decodes visual semantics first and then injects semantically gated detail residuals from the vision tokenizer to refine high-frequency content. Experiments on popular benchmarks demonstrate that Cheers matches or surpasses advanced UMMs in both visual understanding and generation. Cheers also achieves 4x token compression, enabling more efficient high-resolution image encoding and generation. Notably, Cheers outperforms the Tar-1.5B on the popular benchmarks GenEval and MMBench, while requiring only 20% of the training cost, indicating effective and efficient (i.e., 4x token compression) unified multimodal modeling. We will release all code and data for future research.

cs.CV↗

arXiv2026

MARA: A Multimodal Adaptive Retrieval-Augmented Framework for Document Question Answering

Retrieval-based multimodal document QA aims to identify and integrate relevant information from visually rich documents with complex multimodal structures. While retrieval-augmented generation (RAG) has shown strong performance in text-based QA, its extensions to multimodal documents remain underexplored and face significant limitations. Specifically, current approaches rely on query-agnostic document representations that overlook salient content and use static top-k evidence selection, which fails to adapt to the uncertain distribution of relevant information. To address these limitations, we propose the Multimodal Adaptive Retrieval-Augmented (MARA) framework, which introduces query-adaptive mechanisms to both retrieval and generation. MARA consists of two components: a Query-Aligned Region Encoder that builds multi-level document representations and reweights them based on query relevance to improve retrieval precision; and a Self-Reflective Evidence Controller that monitors evidence sufficiency during generation and adaptively incorporates content from lower-ranked sources using a sliding-window strategy. Experiments on six multimodal QA benchmarks demonstrate that MARA consistently improves retrieval relevance and answer quality over existing SOTA method.

cs.IR↗

arXiv2025

RAPTOR: Real-Time High-Resolution UAV Video Prediction with Efficient Video Attention

Video prediction is plagued by a fundamental trilemma: achieving high-resolution and perceptual quality typically comes at the cost of real-time speed, hindering its use in latency-critical applications. This challenge is most acute for autonomous UAVs in dense urban environments, where foreseeing events from high-resolution imagery is non-negotiable for safety. Existing methods, reliant on iterative generation (diffusion, autoregressive models) or quadratic-complexity attention, fail to meet these stringent demands on edge hardware. To break this long-standing trade-off, we introduce RAPTOR, a video prediction architecture that achieves real-time, high-resolution performance. RAPTOR's single-pass design avoids the error accumulation and latency of iterative approaches. Its core innovation is Efficient Video Attention (EVA), a novel translator module that factorizes spatiotemporal modeling. Instead of processing flattened spacetime tokens with $O((ST)^2)$ or $O(ST)$ complexity, EVA alternates operations along the spatial (S) and temporal (T) axes. This factorization reduces the time complexity to $O(S + T)$ and memory complexity to $O(max(S, T))$, enabling global context modeling at $512^2$ resolution and beyond, operating directly on dense feature maps with a patch-free design. Complementing this architecture is a 3-stage training curriculum that progressively refines predictions from coarse structure to sharp, temporally coherent details. Experiments show RAPTOR is the first predictor to exceed 30 FPS on a Jetson AGX Orin for $512^2$ video, setting a new state-of-the-art on UAVid, KTH, and a custom high-resolution dataset in PSNR, SSIM, and LPIPS. Critically, RAPTOR boosts the mission success rate in a real-world UAV navigation task by 18%, paving the way for safer and more anticipatory embodied agents.

cs.CV↗

arXiv2025

GeoViS: Geospatially Rewarded Visual Search for Remote Sensing Visual Grounding

Recent advances in multimodal large language models(MLLMs) have led to remarkable progress in visual grounding, enabling fine-grained cross-modal alignment between textual queries and image regions. However, transferring such capabilities to remote sensing imagery remains challenging, as targets are often extremely small within kilometer-scale scenes, and queries typically involve intricate geospatial relations such as relative positions, spatial hierarchies, or contextual dependencies across distant objects. To address these challenges, we propose GeoViS, a Geospatially Rewarded Visual Search framework that reformulates remote sensing visual grounding as a progressive search-and-reasoning process. Rather than directly predicting the target location in a single step, GeoViS actively explores the global image through a tree-structured sequence of visual cues, integrating multimodal perception, spatial reasoning, and reward-guided exploration to refine geospatial hypotheses iteratively. This design enables the model to detect subtle small-scale targets while maintaining holistic scene awareness. Extensive experiments on five remote sensing grounding benchmarks demonstrate that GeoViS achieves precise geospatial understanding and consistently surpasses existing methods across key visual grounding metrics, highlighting its strong cross-domain generalization and interpretability.

cs.CV↗

arXiv2025

LLaVA-UHD v3: Progressive Visual Compression for Efficient Native-Resolution Encoding in MLLMs

Visual encoding followed by token condensing has become the standard architectural paradigm in multi-modal large language models (MLLMs). Many recent MLLMs increasingly favor global native- resolution visual encoding over slice-based methods. To investigate this trend, we systematically compare their behavior on vision-language understanding and attention patterns, revealing that global encoding enhances overall capability but at the expense of greater computational overhead. To address this issue, we present LLaVA-UHD v3, an MLLM centered upon our proposed Progressive Visual Compression (PVC) method, which can be seamlessly integrated into standard Vision Transformer (ViT) to enable efficient native-resolution encoding. The PVC approach consists of two key modules: (i) refined patch embedding, which supports flexible patch-size scaling for fine-grained visual model- ing, (ii) windowed token compression, hierarchically deployed across ViT layers to progressively aggregate local token representations. Jointly modulated by these two modules, a widely pretrained ViT can be reconfigured into an efficient architecture while largely preserving generality. Evaluated across extensive benchmarks, the transformed ViT, termed ViT-UHD, demonstrates competitive performance with MoonViT while reducing TTFT (time-to-first-token) by 2.4x, when developed within an identical MLLM architecture. Building upon ViT-UHD, LLaVA-UHD v3 also achieves competitive performance to Qwen2-VL, while further reducing TTFT by 1.9x. We will release all code and checkpoints to support future research on efficient MLLMs.

cs.CV↗

arXiv2025

ReconViaGen: Towards Accurate Multi-view 3D Object Reconstruction via Generation

Existing multi-view 3D object reconstruction methods heavily rely on sufficient overlap between input views, where occlusions and sparse coverage in practice frequently yield severe reconstruction incompleteness. Recent advancements in diffusion-based 3D generative techniques offer the potential to address these limitations by leveraging learned generative priors to hallucinate invisible parts of objects, thereby generating plausible 3D structures. However, the stochastic nature of the inference process limits the accuracy and reliability of generation results, preventing existing reconstruction frameworks from integrating such 3D generative priors. In this work, we comprehensively analyze the reasons why diffusion-based 3D generative methods fail to achieve high consistency, including (a) the insufficiency in constructing and leveraging cross-view connections when extracting multi-view image features as conditions, and (b) the poor controllability of iterative denoising during local detail generation, which easily leads to plausible but inconsistent fine geometric and texture details with inputs. Accordingly, we propose ReconViaGen to innovatively integrate reconstruction priors into the generative framework and devise several strategies that effectively address these issues. Extensive experiments demonstrate that our ReconViaGen can reconstruct complete and accurate 3D models consistent with input views in both global structure and local details.Project page: https://jiahao620.github.io/reconviagen.

cs.CV↗

arXiv2025

CFVBench: A Comprehensive Video Benchmark for Fine-grained Multimodal Retrieval-Augmented Generation

Multimodal Retrieval-Augmented Generation (MRAG) enables Multimodal Large Language Models (MLLMs) to generate responses with external multimodal evidence, and numerous video-based MRAG benchmarks have been proposed to evaluate model capabilities across retrieval and generation stages. However, existing benchmarks remain limited in modality coverage and format diversity, often focusing on single- or limited-modality tasks, or coarse-grained scene understanding. To address these gaps, we introduce CFVBench, a large-scale, manually verified benchmark constructed from 599 publicly available videos, yielding 5,360 open-ended QA pairs. CFVBench spans high-density formats and domains such as chart-heavy reports, news broadcasts, and software tutorials, requiring models to retrieve and reason over long temporal video spans while maintaining fine-grained multimodal information. Using CFVBench, we systematically evaluate 7 retrieval methods and 14 widely-used MLLMs, revealing a critical bottleneck: current models (even GPT5 or Gemini) struggle to capture transient yet essential fine-grained multimodal details. To mitigate this, we propose Adaptive Visual Refinement (AVR), a simple yet effective framework that adaptively increases frame sampling density and selectively invokes external tools when necessary. Experiments show that AVR consistently enhances fine-grained multimodal comprehension and improves performance across all evaluated MLLMs

cs.CL↗

arXiv2025

VC-Agent: An Interactive Agent for Customized Video Dataset Collection

Facing scaling laws, video data from the internet becomes increasingly important. However, collecting extensive videos that meet specific needs is extremely labor-intensive and time-consuming. In this work, we study the way to expedite this collection process and propose VC-Agent, the first interactive agent that is able to understand users' queries and feedback, and accordingly retrieve/scale up relevant video clips with minimal user input. Specifically, considering the user interface, our agent defines various user-friendly ways for the user to specify requirements based on textual descriptions and confirmations. As for agent functions, we leverage existing multi-modal large language models to connect the user's requirements with the video content. More importantly, we propose two novel filtering policies that can be updated when user interaction is continually performed. Finally, we provide a new benchmark for personalized video dataset collection, and carefully conduct the user study to verify our agent's usage in various real scenarios. Extensive experiments demonstrate the effectiveness and efficiency of our agent for customized video dataset collection. Project page: https://allenyidan.github.io/vcagent_page/.

cs.AI↗

arXiv2025

Community Bail Fund Systems: Fluid Limits and Approximations

Community bail funds (CBFs) assist individuals who have been arrested and cannot afford bail, preventing unnecessary pretrial incarceration along with its harmful or sometimes fatal consequences. By posting bail, CBFs allow defendants to stay at home and maintain their livelihoods until trial. This paper introduces new stochastic models that combine queueing theory with classic insurance risk models to capture the dynamics of the remaining funds in a CBF. We first analyze a model where all bail requests are accepted. Although the remaining fund balance can go negative, this model provides insight for CBFs that are not financially constrained. We then apply the Skorokhod map to make sure the CBF balance does not go negative and show that the Skorokhod map produces a model where requests are partially fulfilled. Finally, we analyze a model where bail requests can be blocked if there is not enough money to satisfy the request upon arrival. Although the blocking model prevents the CBF from being negative, the blocking feature gives rise to new analytical challenges for a direct stochastic analysis. Thus, we prove a functional law of large numbers or a fluid limit for the blocking model and show that the fluid limit is a distributed delay equation. We assess the quality of our fluid limit via simulation and show that the fluid limit accurately describes the large-scale stochastic dynamics of the CBF. Finally, we prove stochastic ordering results for the CBF processes we analyze.

math.PR↗

arXiv2025

ReSeDis: A Dataset for Referring-based Object Search across Large-Scale Image Collections

Large-scale visual search engines are expected to solve a dual problem at once: (i) locate every image that truly contains the object described by a sentence and (ii) identify the object's bounding box or exact pixels within each hit. Existing techniques address only one side of this challenge. Visual grounding yields tight boxes and masks but rests on the unrealistic assumption that the object is present in every test image, producing a flood of false alarms when applied to web-scale collections. Text-to-image retrieval excels at sifting through massive databases to rank relevant images, yet it stops at whole-image matches and offers no fine-grained localization. We introduce Referring Search and Discovery (ReSeDis), the first task that unifies corpus-level retrieval with pixel-level grounding. Given a free-form description, a ReSeDis model must decide whether the queried object appears in each image and, if so, where it is, returning bounding boxes or segmentation masks. To enable rigorous study, we curate a benchmark in which every description maps uniquely to object instances scattered across a large, diverse corpus, eliminating unintended matches. We further design a task-specific metric that jointly scores retrieval recall and localization precision. Finally, we provide a straightforward zero-shot baseline using a frozen vision-language model, revealing significant headroom for future study. ReSeDis offers a realistic, end-to-end testbed for building the next generation of robust and scalable multimodal search systems.

cs.CV↗

arXiv2025

Soliton resolution, asymptotic stability and Painlevé transcendents in the combined Wadati-Konno-Ichikawa and short-pulse equation

In this paper, we develop a Riemann-Hilbert (RH) approach to the Cauchy problem for the combined Wadati-Konno-Ichikawa and short-pulse (WKI-SP) equation. The solution of the Cauchy problem is first expressed in terms of the solution of a RH problem with direct scattering transform based on the Lax pair. Further through a series of deformations to the RH problem by using the $\bar{\partial}$-generalization of Deift-Zhou steepest descent method, we obtain the long-time asymptotic approximations to the solution of the WKI-SP equation under a new scale $(y,t)$ in three kinds of space-time regions. The first asymptotic result from the space-time regions $ ξ:=y/t <-2\sqrt{3αβ}, αβ>0$ and $|ξ|<\infty,αβ<0$ with saddle points on $\mathbb{R}$, is characterized with solitons and soliton-radiation interaction with residual error $\mathcal{O}(t^{-3/4})$. The second asymptotic result from the region $ ξ>-2\sqrt{3αβ}, αβ>0$ without saddle point on $\mathbb{R}$, is characterized with modulation-solitons with residual error $\mathcal{O}(t^{-1})$; These two results above are a verification of the soliton resolution conjecture for the WKI-SP equation. The third asymptotic result from a transition region $ξ\approx -2\sqrt{3αβ},αβ>0$ can be expressed in terms of the solution of the Painlevé \uppercase\expandafter{\romannumeral2} equation with error $\mathcal{O}(t^{-1/2})$. This is a new phenomena that the long-time asymptotics for the solution to the Cauchy problem of the WKI equation and SP equation don't possesses.

math-ph↗

arXiv2025

Unveiling Language-Specific Features in Large Language Models via Sparse Autoencoders

The mechanisms behind multilingual capabilities in Large Language Models (LLMs) have been examined using neuron-based or internal-activation-based methods. However, these methods often face challenges such as superposition and layer-wise activation variance, which limit their reliability. Sparse Autoencoders (SAEs) offer a more nuanced analysis by decomposing the activations of LLMs into a sparse linear combination of SAE features. We introduce a novel metric to assess the monolinguality of features obtained from SAEs, discovering that some features are strongly related to specific languages. Additionally, we show that ablating these SAE features only significantly reduces abilities in one language of LLMs, leaving others almost unaffected. Interestingly, we find some languages have multiple synergistic SAE features, and ablating them together yields greater improvement than ablating individually. Moreover, we leverage these SAE-derived language-specific features to enhance steering vectors, achieving control over the language generated by LLMs. The code is publicly available at https://github.com/Aatrox103/multilingual-llm-features.

cs.CL↗