arXiv subjects

Yuxuan Yang

Publications and source records attributed to Yuxuan Yang.

At least 19 recordsLinked to original sources

arXiv2026

Mind the Couch! Eliciting MLLM Reasoning in Interior Design via Weak-to-Strong Task Vector Injection

Multimodal Large Language Models (MLLMs) have demonstrated great performance, yet they often suffer from severe modality misalignment when confronted with densely constrained spaces for interior design. Due to the loss of high-frequency local topological details and fine-grained aesthetic shifts during visual encoding, existing MLLMs frequently hallucinate, yielding physical spatial collisions and visual aesthetic dissonance. To address this, we propose Dual-prior Activation Residual Task-vectors Injection mechanism (DART-I) for MLLMs. It shifts the paradigm from lossy text-prompting to direct latent intervention, utilizing weak-to-strong deterministic rules to anchor the causal reasoning of MLLMs for interior design. Specifically, DART-I operates in three steps: it first explicitly extracts continuous spatial distance and color typography features from images using extremely lightweight weak experts; subsequently, it transforms these deterministic priors into directional task vectors via a linear projection network; these vectors are dynamically injected as residual terms into the latent space of the frozen MLLMs, steering MLLMs towards precise reasoning for interior design. Stepping outside the conventional paradigms, our method achieves precise reasoning without fine-tuning the MLLMs, effectively bypassing expensive computational costs and catastrophic forgetting. Extensive experiments on various benchmarks demonstrate the effectiveness and advantages of DART-I.

cs.MM↗

arXiv2026

Inversion Diameter of Planar Graphs

Given an oriented graph $\vec{G}$ and a subset of vertices $X \subseteq V(\vec{G})$, the \emph{inversion} of $X$ is the operation that reverses the orientation of every arc with both endpoints in $X$. For a simple graph $G$, the inversion diameter $\operatorname{diam}(I(G))$ is the maximum distance between two orientations of $G$ under inversions of vertex sets. We prove the sharp bound \[ \operatorname{diam}(I(G))\le 2\chi_a(G)-2, \] where $\chi_a(G)$ is the acyclic chromatic number. Consequently, every planar graph has inversion diameter at most $8$, improving the previously known bound $12$. Using strong-degeneracy arguments, we also obtain upper bounds $7$, $5$, and $4$ for planar graphs of girth at least $4$, $5$, and $6$, respectively.

math.CO↗

arXiv2026

Towards General Language-Conditioned Latent Safety Filters

Robot policies are becoming increasingly general, with vision-language-action (VLA) models enabling a single policy to execute diverse tasks specified in natural language. Safe deployment, however, requires adapting not only to new tasks but also to varying safety requirements across users, environments, and applications. Existing safety filters remain largely constraint-specific and thus must be redesigned or relearned when safety requirements change. In this paper, we investigate language-conditioned safety filtering, in which a Hamilton-Jacobi safety actor and critic are conditioned on language-specified constraints. We evaluate this formulation across pick-and-place, table-wiping, and block-stacking tasks in the vision-based setting, examining its ability to enforce language-specified constraints and transfer to unseen constraint instances within the evaluated constraint families. Our experiments provide evidence that language-conditioned safety filters reduce constraint violations and exhibit partial transfer to unseen constraint instances.

cs.RO↗

arXiv2026

HARD-KV: Head-Adaptive Regularization for Decoding-time KV Compression

Long-context LLM inference faces a fundamental conflict: head-adaptive compression algorithms (e.g., Top-$p$ nucleus sampling) offer superior accuracy by dynamically fluctuating memory budgets, yet modern inference engines (e.g., vLLM) demand rigid, static memory patterns to leverage CUDA Graphs and PagedAttention. We resolve this ``Static-Dynamic'' mismatch with HARD-KV, a unified framework that that bridges dynamic selection with rigid system constraints. HARD-KV introduces a Cascade Cache hierarchy, managing the token lifecycle across dense, sparse, and condensed tiers. Crucially, we propose a Logits Calibration mechanism that normalizes diverse importance metrics into a unified probability space, enabling consistent Top-$p$ budgeting across heterogeneous heads. To bridge the efficiency gap, we offer a system-level solution, which rewrites fragmented, dynamic indices into contiguous physical layouts compatible with high-performance inference engine. Extensive experiments on math-reasoning benchmarks (AIME, U-Math) verify that HARD-KV achieves up to 2$\times$ throughput improvement over static baselines while maintaining high-fidelity generation in 10k+ token scenarios. Code is available at https://github.com/SuDIS-ZJU/HARDInfer.

cs.LG↗

arXiv2026

DiARC: Distinguishing Positive and Negative Samples Helps Improving ARC-like Reasoning Ability of Large Language Models

The Abstraction and Reasoning Corpus (ARC) contains tasks that require summarizing patterns from limited grid samples and predicting output grids. Recently, many large language model based approaches have attempted to transform it into a text-based reasoning task. However, methods based on open-source models have generally yielded unsatisfactory results, while those relying on closed-source models are too costly. Current efforts mainly focus on data augmentation, constructing ARC-like data for more comprehensive supervised fine-tuning. In this work, we argue that solving ARC-like problems requires not only positive sample supervision but also the ability to improve model reasoning by distinguishing negative samples. To this end, we draw on the idea of preference alignment and propose DiARC, a method that constructs preference pairs to enable the model to distinguish between them. Specifically, we propose three ways to construct negative samples, including output-level visual transformations, DSL-level rule inversion, and task-specific rule editing. The resulting negative samples provide informative near-miss alternatives while keeping the observed demonstrations unchanged. Experimental results across multiple ARC-like benchmarks show that DiARC consistently improves performance over baseline models. The code is released at https://github.com/szu-tera/DiARC.

cs.CL↗

arXiv2026

Potential-Guided Flow Matching for Vision-Language-Action Policy Improvement

Large vision-language-action (VLA) policies are increasingly trained as conditional generative models over action chunks. Yet deployment produces mixed-quality experience-successful demonstrations, partial completions, recoverable mistakes, and failures-that is difficult to use with standard imitation. Full behavior cloning (BC) imitates failures, filtered BC discards useful sub-trajectories, and offline reinforcement learning adds a large critic. We introduce ForesightFlow, a self-guided flow-matching policy that augments each generated action chunk with a learned success-potential trajectory. The same flow proposes and scores candidate actions, enabling best-of-$K$ inference without an external critic. The key issue is that policy improvement and value calibration require different supervision: advantage weighting should emphasize high-quality actions, but applying the same weights to potential coordinates suppresses failure gradients and creates overconfident scores. We address this with decoupled advantage-weighted flow matching, applying exponentiated advantage weights only to action velocities while training potential velocities uniformly. We further derive a one-step boundary estimator for conditional flow matching, allowing advantage computation with a single stop-gradient forward pass. Across five BEHAVIOR-1K simulation tasks and five real-world bimanual tasks, ForesightFlow improves over imitation baselines, matches the strongest separate-critic baseline in simulation success, improves real-world success, and reduces training compute by $38\%$. Ablations show that decoupling prevents value hallucination, the one-step estimator preserves candidate-ranking fidelity, and self-guided sampling improves long-horizon execution.

cs.RO↗

arXiv2026

SMAC: Spatial-Modal Joint Modeling and Adaptive Representation Collapse for Multimodal Object Tracking

Multimodal multi-object tracking (MOT) under complex illumination remains challenging due to insufficient joint modeling of spatial and modal features and the limited adaptability of fixed fusion strategies. To address these issues, this paper proposes a spatial-modal convolution fusion and distillation-prompt-based multimodal MOT framework. A spatial-modal fusion backbone is first constructed, where a Basic module performs spatial feature extraction and modal interaction via decoupled 3D convolution, while a Mixed module models nonlinear cross-modal correlations through amplitude-phase decomposition. In addition, a representation collapse network is designed for adaptive multimodal fusion. A Distillation Prompt Guidance (DPG) module generates dynamic modal weights under teacher supervision, and a Global Modal Difference Aggregation (GMDA) module preserves discriminative information during multimodal representation collapse. Extensive experiments on the UniRTL dataset demonstrate the effectiveness of the proposed method. The proposed tracker achieves 63.31 HOTA and 79.21 MOTA on the RNT modality, outperforming several state-of-the-art methods while maintaining favorable inference efficiency. The source code and pretrained models are publicly available at https://github.com/QitaiSun/SMAC.

eess.IV↗

arXiv2026

Optimal embeddings for maximal orders of central simple algebras of degree 3 over number fields

Let $B$ be a central simple algebra of degree 3 over a number field $F$ and $K/F$ be a finite extension of degree 3. For an order $S$ of $K$, we determine exactly when $S$ cannot be optimally embedded into all maximal orders of $B$. Moreover, we further determine exactly when $S$ can be optimally embedded into $\frac{1}{3}$ isomorphism classes of maximal orders of $B$ and $\frac{2}{3}$ isomorphism classes of maximal orders of $B$ in the rest of cases.

math.NT↗

arXiv2026

CombinationTS: A Modular Framework for Understanding Time-Series Forecasting Models

Recent progress in time-series forecasting has led to rapidly increasing architectural complexity, yet many reported State-of-the-Art gains are statistically fragile or misattributed. We argue that progress requires a shift from model selection to modular attribution, identifying which components truly drive performance. We propose CombinationTS, a self-contained probabilistic evaluation framework that decomposes forecasting models into orthogonal modules--Input Transformation, Embedding, Encoder, Decoder, and Output Transformation--and evaluates them under a shared evaluation condition space. By quantifying each component via marginalized performance ($\mu$) and stability ($\sigma$), CombinationTS enables robust attribution beyond fragile point estimates. Through large-scale paired evaluation, we uncover the Identity Paradox: once the data view (Embedding) is well-designed, a parameter-free Identity Encoder often matches or outperforms complex backbones. We further show that explicit structural priors introduced via Input Transformations yield a more favorable performance-stability trade-off than increasing Encoder complexity, establishing a principled baseline for architectural necessity.

cs.LG↗

arXiv2026

SP-GCRL: Influence Maximization on Incomplete Social Graphs

Influence maximization (IM) in real platforms is challenged by incomplete, noisy social graphs and non-stationary diffusion dynamics. We propose SP-GCRL, a social-propagation-aware graph contrastive reinforcement learning framework that learns end-to-end seed selection under partial observability.We first introduce a social-propagation-aware nonlinear diffusion function to model reinforcement/diminishing effects and probability drift under repeated exposure; we then construct dual structural views and perform contrastive learning to obtain node representations robust to missing edges and weak ties, while replacing expensive strategy metrics with a GAT-based regression surrogate to improve efficiency and scalability; finally, we use DDQN to learn an end-to-end seed selection policy on top of these representations. Experiments on multiple real-world networks show that SP-GCRL achieves significant gains over heuristic and learning-based baselines across budgets and topologies, while maintaining strong large-scale scalability.

cs.SI↗

arXiv2026

Evaluating Uplift Modeling under Structural Biases: Insights into Metric Stability and Model Robustness

In personalized marketing, uplift models estimate the incremental effect of an intervention by modeling how customer behavior would change under alternative treatments using counterfactual analysis. However, real-world marketing data often exhibit various biases, such as selection bias, spillover effects, measurement error, and unobserved confounding. These biases can adversely affect both the accuracy of uplift estimation and the validity of evaluation metrics. Despite the importance of bias-aware assessment, there remains a lack of systematic studies evaluating how different models and metrics perform under such biased conditions. To bridge this gap, we design a systematic benchmarking framework. Unlike standard predictive tasks, real-world uplift datasets inherently lack counterfactual ground truth. This limitation renders the direct validation of evaluation metrics infeasible and prevents the precise quantification of biases. Therefore, a semi-synthetic approach serves as a critical enabler for systematic benchmarking. This approach effectively bridges the gap by retaining real-world feature dependencies while providing the ground truth needed to isolate structural biases. Our investigations reveal that (i) uplift targeting and prediction can manifest as distinct objectives, where proficiency in one does not ensure efficacy in the other; (ii) while many models exhibit inconsistent performance under diverse biases, TARNet shows notable robustness, providing insights for subsequent model design; (iii) the stability of evaluation metrics is linked to their mathematical alignment with the ATE, suggesting that ATE-approximating metrics yield more consistent model rankings under structural data imperfections. These findings suggest the need for more robust uplift models and evaluation metrics under real-world data imperfections.

cs.LG↗

arXiv2026

Hierarchy of extreme-event predictability in turbulence revealed by machine learning

Extreme-event predictability in turbulence is strongly state dependent, yet event-by-event predictability horizons are difficult to quantify without access to governing equations or costly perturbation ensembles. Here we train an autoregressive conditional diffusion model on direct numerical simulations of the two-dimensional Kolmogorov flow and use a CRPS-based skill score to define an event-wise predictability horizon. Enstrophy extremes exhibit a pronounced hierarchy: forecast skill persists from $\approx 1$ to $> 4$ Lyapunov times across events. Spectral filtering shows that these horizons are controlled predominantly by large-scale structures. Extremes are preceded by intense strain cores organizing quadrupolar vortex packets, whose lifetime sharply separates long- from short-horizon events. These results identify coherent-structure persistence as a governing mechanism for the predictability of turbulence extremes and provide a data-driven route to diagnose predictability limits from observations.

nlin.CD↗

arXiv2026

Selfish Cooperation Towards Low-Altitude Economy: Integrated Multi-Service Deployment with Resilient Federated Reinforcement Learning

The low-altitude economy (LAE) is a rapidly emerging paradigm that builds a service-centric economic ecosystem through large-scale and sustainable uncrewed aerial vehicle (UAV)-enabled service provisioning, reflecting the transition of the 6G era from technological advancement toward commercial deployment. The significant market potential of LAE attracts an increasing number of service providers (SPs), resulting in intensified competition in service deployment. In this paper, we study a realistic LAE scenario in which multiple SPs dynamically deploy UAVs to deliver multiple services to user hotspots, aiming to jointly optimize communication and computation resource allocation. To resolve deployment competition among SPs, an authenticity-guaranteed auction mechanism is designed, and game-theoretic analysis is conducted to establish the solvability of the proposed resource allocation problem. Furthermore, a resilient federated reinforcement learning (FRL)-based solution is developed with strong fault tolerance, effectively countering transmission errors and malicious competition while facilitating potential cooperation among self-interested SPs. Simulation results demonstrate that the proposed approach significantly improves service performance and robustness compared with baseline methods, providing a practical and scalable solution for competitive LAE service deployment.

cs.NI↗

arXiv2026

Adaptive Personalized Federated Reinforcement Learning for RIS-Assisted Aerial Relays in SAGINs with Fluid Antennas

Space-air-ground integrated networks (SAGINs) interconnect satellites, uncrewed aerial vehicles (UAVs), and ground devices to enable flexible and ubiquitous wireless services. The integration of reconfigurable intelligent surfaces (RISs) and fluid antenna systems (FASs) further enhances radio environment controllability. However, the tight integration of cross-layer facilities and radio enhancement technologies leads to pronounced environmental dynamics and heterogeneity, posing fundamental challenges for system modeling and optimization in large-scale SAGINs. This paper investigates a SAGIN in which low Earth orbit (LEO) satellite constellations communicate with multiple ground hotspots via RIS-assisted UAV relays, serving both FAS-equipped and conventional users. A system model is developed that explicitly captures satellite mobility, UAV trajectories, RIS phase control, and heterogeneous user reception capabilities. Accordingly, a multi-hotspot downlink rate maximization problem is studied, whose solvability is analyzed through a hierarchical Stackelberg game. To address heterogeneous and time-varying multi-hotspot environments, an adaptive personalized federated reinforcement learning (FRL) algorithm is proposed for adaptive optimization of UAV trajectories and RIS phase controls. Simulation results demonstrate superior performance and validate the effectiveness of personalization in dynamic heterogeneous SAGIN scenarios.

cs.NI↗

arXiv2026

Design-MLLM: A Reinforcement Alignment Framework for Verifiable and Aesthetic Interior Design

Interior design is a requirements-to-visual-plan generation process that must simultaneously satisfy verifiable spatial feasibility and comparative aesthetic preferences. While recent multimodal large language models (MLLMs) offer a unified foundation for interpreting user intent and producing design rationales, our empirical analysis reveals a persistent contradiction in real-world deployment: MLLMs often produce layouts that are unbuildable and aesthetically inconsistent. These findings indicate that simply adding in-domain text is insufficient; effective interior design requires an alignment mechanism that separates hard constraints from soft preferences and coordinates them during optimization. To address this, we propose Design-MLLM, a reinforcement alignment framework that optimizes a feasibility-first preference objective via a dual-branch, aesthetic-oriented reward. Specifically, Design-MLLM (i) explicitly evaluates spatial feasibility using programmatic constraint checks, (ii) assesses aesthetic preference only among feasible candidates to avoid visually appealing but unexecutable shortcuts, and (iii) performs group-relative optimization to obtain stable preference signals. Through this process, Design-MLLM learns a controllable policy that consistently selects and generates solutions that are both executable and aesthetically coherent, rather than occasionally producing visually appealing but infeasible designs. Extensive experiments on various benchmark datasets demonstrate the advantages of Design-MLLM.

cs.MM↗

arXiv2026

Hepato-LLaVA: An Expert MLLM with Sparse Topo-Pack Attention for Hepatocellular Pathology Analysis on Whole Slide Images

Hepatocellular Carcinoma diagnosis relies heavily on the interpretation of gigapixel Whole Slide Images. However, current computational approaches are constrained by fixed-resolution processing mechanisms and inefficient feature aggregation, which inevitably lead to either severe information loss or high feature redundancy. To address these challenges, we propose Hepato-LLaVA, a specialized Multi-modal Large Language Model designed for fine-grained hepatocellular pathology analysis. We introduce a novel Sparse Topo-Pack Attention mechanism that explicitly models 2D tissue topology. This mechanism effectively aggregates local diagnostic evidence into semantic summary tokens while preserving global context. Furthermore, to overcome the lack of multi-scale data, we present HepatoPathoVQA, a clinically grounded dataset comprising 33K hierarchically structured question-answer pairs validated by expert pathologists. Our experiments demonstrate that Hepato-LLaVA achieves state-of-the-art performance on HCC diagnosis and captioning tasks, significantly outperforming existing methods. Our code and implementation details are available at https://pris-cv.github.io/Hepto-LLaVA/.

cs.CV↗

arXiv2025

Single-View Shape Completion for Robotic Grasping in Clutter

In vision-based robot manipulation, a single camera view can only capture one side of objects of interest, with additional occlusions in cluttered scenes further restricting visibility. As a result, the observed geometry is incomplete, and grasp estimation algorithms perform suboptimally. To address this limitation, we leverage diffusion models to perform category-level 3D shape completion from partial depth observations obtained from a single view, reconstructing complete object geometries to provide richer context for grasp planning. Our method focuses on common household items with diverse geometries, generating full 3D shapes that serve as input to downstream grasp inference networks. Unlike prior work, which primarily considers isolated objects or minimal clutter, we evaluate shape completion and grasping in realistic clutter scenarios with household objects. In preliminary evaluations on a cluttered scene, our approach consistently results in better grasp success rates than a naive baseline without shape completion by 23% and over a recent state of the art shape completion approach by 19%. Our code is available at https://amm.aass.oru.se/shape-completion-grasping/.

cs.RO↗

arXiv2025

Maximal orders optimal embedding of central simple algebras over number fields

Given a number field $F$ and $R$ be the ring of integers of $F$, the problem of embedding a field extension $K/F$ into a central simple algebra $B$ is classical. This paper proves that when the central simple algebra has degree $p$, the $R$-order $S\subset K$ can be optimal embedded into all maximal $R$-orders $O\subset B$, unless satisfies the optimal selectivity condition.

math.NT↗