arXiv subjects

Han Zhang

Publications and source records attributed to Han Zhang.

At least 127 records · Page 7Linked to original sources

arXiv2026

A Synergistic Approach: Dynamics-AI Ensemble in Tropical Cyclone Forecasting

This study addresses a critical challenge in AI-based weather forecasting by developing an AI-driven optimized ensemble forecast system using Orthogonal Conditional Nonlinear Optimal Perturbations (O-CNOPs). The system bridges the gap between computational efficiency and dynamic consistency in tropical cyclone (TC) forecasting. Unlike conventional ensembles limited by computational costs or AI ensembles constrained by inadequate perturbation methods, O-CNOPs generate dynamically optimized perturbations that capture fast-growing errors of FuXi model while maintaining plausibility. The key innovation lies in producing orthogonal perturbations that respect FuXi nonlinear dynamics, yielding structures reflecting dominant dynamical controls and physically interpretable probabilistic forecasts. Demonstrating superior deterministic and probabilistic skills over the operational Integrated Forecasting System Ensemble Prediction System, this work establishes a new paradigm combining AI computational advantages with rigorous dynamical constraints. Success in TC track forecasting paves the way for reliable ensemble forecasts of other high-impact weather systems, marking a major step toward operational AI-based ensemble forecasting.

physics.ao-ph↗

arXiv2026

Khintchine dichotomy and Schmidt estimates for self-similar measures on $\mathbb{R}^d$

We extend the classical theorems of Khintchine and Schmidt in metric Diophantine approximation to the context of self-similar measures on $\mathbb{R}^d$. For this, we establish effective equidistribution of associated random walks on $\text{SL}_{d+1}(\mathbb{R})/\text{SL}_{d+1}(\mathbb{Z})$. This generalizes our previous work which requires $d=1$ and restricts Schmidt-type counting estimates to approximation functions which decay fast enough. Novel techniques include a bootstrap scheme for the associated random walks despite algebraic obstructions, and a refined treatment of Dani's correspondence. Along the way, we also establish non-concentration properties of self-similar measures near algebraic subvarieties of $\mathbb{R}^d$.

math.DS↗

arXiv2026

Lifelong Scalable Multi-Agent Realistic Testbed and A Comprehensive Study on Design Choices in Lifelong AGV Fleet Management Systems

We present Lifelong Scalable Multi-Agent Realistic Testbed (LSMART), an open-source simulator to evaluate any Multi-Agent Path Finding (MAPF) algorithm in a Fleet Management System (FMS) with Automated Guided Vehicles (AGVs). MAPF aims to move a group of agents from their corresponding starting locations to their goals. Lifelong MAPF (LMAPF) is a variant of MAPF that continuously assigns new goals for agents to reach. LMAPF applications, such as autonomous warehouses, often require a centralized, lifelong system to coordinate the movement of a fleet of robots, typically AGVs. However, existing works on MAPF and LMAPF often assume simplified kinodynamic models, such as pebble motion, as well as perfect execution and communication for AGVs. Prior work has presented SMART, a software capable of evaluating any MAPF algorithms while considering agent kinodynamics, communication delays, and execution uncertainties. However, SMART is designed for MAPF, not LMAPF. Generalizing SMART to an FMS requires many more design choices. First, an FMS parallelizes planning and execution, raising the question of when to plan. Second, given planners with varying optimality and differing agent-model assumptions, one must decide how to plan. Third, when the planner fails to return valid solutions, the system must determine how to recover. In this paper, we first present LSMART, an open-source simulator that incorporates all these considerations to evaluate any MAPF algorithms in an FMS. We then provide experiment results based on state-of-the-art methods for each design choice, offering guidance on how to effectively design centralized lifelong AGV Fleet Management Systems. LSMART is available at https://smart-mapf.github.io/lifelong-smart.

cs.RO↗

arXiv2026

MTFM: A Scalable and Alignment-free Foundation Model for Industrial Recommendation in Meituan

Industrial recommendation systems typically involve multiple scenarios, yet existing cross-domain (CDR) and multi-scenario (MSR) methods often require prohibitive resources and strict input alignment, limiting their extensibility. We propose MTFM (Meituan Foundation Model for Recommendation), a transformer-based framework that addresses these challenges. Instead of pre-aligning inputs, MTFM transforms cross-domain data into heterogeneous tokens, capturing multi-scenario knowledge in an alignment-free manner. To enhance efficiency, we first introduce a multi-scenario user-level sample aggregation that significantly enhances training throughput by reducing the total number of instances. We further integrate Grouped-Query Attention and a customized Hybrid Target Attention to minimize memory usage and computational complexity. Furthermore, we implement various system-level optimizations, such as kernel fusion and the elimination of CPU-GPU blocking, to further enhance both training and inference throughput. Offline and online experiments validate the effectiveness of MTFM, demonstrating that significant performance gains are achieved by scaling both model capacity and multi-scenario training data.

cs.IR↗

arXiv2026

Deconstructing Positional Information: From Attention Logits to Training Biases

Positional encodings enable Transformers to incorporate sequential information, yet their theoretical understanding remains limited to two properties: distance attenuation and translation invariance. Because natural language lacks purely positional data, the interplay between positional and semantic information is still underexplored. We address this gap by deconstructing the attention-logit computation and providing a structured analysis of positional encodings, categorizing them into additive and multiplicative forms. The differing properties of these forms lead to distinct mechanisms for capturing positional information. To probe this difference, we design a synthetic task that explicitly requires strong integration of positional and semantic cues. As predicted, multiplicative encodings achieve a clear performance advantage on this task. Moreover, our evaluation reveals a hidden training bias: an information aggregation effect in shallow layers that we term the single-head deposit pattern. Through ablation studies and theoretical analysis, we proved that this phenomenon is inherent in multiplicative encodings. These findings deepen the understanding of positional encodings and call for further study of their training dynamics.

cs.LG↗

arXiv2026

STELAR-VISION: Self-Topology-Aware Efficient Learning for Aligned Reasoning in Vision

Vision-language models (VLMs) have made significant strides in reasoning, yet they often struggle with complex multimodal tasks and tend to generate overly verbose outputs. A key limitation is their reliance on chain-of-thought (CoT) reasoning, despite many tasks benefiting from alternative topologies like trees or graphs. To address this, we introduce STELAR-Vision, a training framework for topology-aware reasoning. At its core is TopoAug, a synthetic data pipeline that enriches training with diverse topological structures. Using supervised fine-tuning and reinforcement learning, we post-train Qwen2VL models with both accuracy and efficiency in mind. Additionally, we propose Frugal Learning, which reduces output length with minimal accuracy loss. On MATH-V and VLM-S2H, STELAR-Vision improves accuracy by 9.7% over its base model and surpasses the larger Qwen2VL-72B-Instruct by 7.3%. On five out-of-distribution benchmarks, it outperforms Phi-4-Multimodal-Instruct by up to 28.4% and LLaMA-3.2-11B-Vision-Instruct by up to 13.2%, demonstrating strong generalization. Compared to Chain-Only training, our approach achieves 4.3% higher overall accuracy on in-distribution datasets and consistently outperforms across all OOD benchmarks.

cs.AI↗

arXiv2026

Foundation Model-Aided Hierarchical Deep Reinforcement Learning for Blockage-Aware Link in RIS-Assisted Networks

Reconfigurable intelligent surface (RIS) technology has the potential to significantly enhance the spectral efficiency (SE) of 6G wireless networks. However, practical deployment remains constrained by challenges in accurate channel estimation and control optimization under dynamic conditions. This paper presents a foundation model-aided hierarchical deep reinforcement learning (FM-HDRL) framework designed for joint beamforming and phase-shift optimization in RIS-assisted wireless networks. To implement this, we first fine-tune a pre-trained large wireless model (LWM) to translate raw channel data into low-dimensional, context-aware channel state information (CSI) embeddings. Next, these embeddings are combined with user location information and blockage status to select the optimal communication path. The resulting features are then fed into an HDRL model, assumed to be implemented at a centralized controller, which jointly optimizes the base station (BS) beamforming vectors and the RIS phase-shift configurations to maximize SE. Simulation results demonstrate that the proposed FM-HDRL framework consistently outperforms baseline methods in terms of convergence speed, spectral efficiency, and scalability. According to the simulation results, our proposed method improves 7.82% SE compared to the FM-aided deep reinforcement learning (FM-DRL) approach and a substantial enhancement of about 48.66% relative to the beam sweeping approach.

eess.SP↗

arXiv2026

MAMBO-G: Magnitude-Aware Mitigation for Boosted Guidance

High-fidelity text-to-image and text-to-video generation typically relies on Classifier-Free Guidance (CFG), but achieving optimal results often demands computationally expensive sampling schedules. In this work, we propose MAMBO-G, a training-free acceleration framework that significantly reduces computational cost by dynamically optimizing guidance magnitudes. We observe that standard CFG schedules are inefficient, applying disproportionately large updates in early steps that hinder convergence speed. MAMBO-G mitigates this by modulating the guidance scale based on the update-to-prediction magnitude ratio, effectively stabilizing the trajectory and enabling rapid convergence. This efficiency is particularly vital for resource-intensive tasks like video generation. Our method serves as a universal plug-and-play accelerator, achieving up to 3x speedup on Stable Diffusion v3.5 (SD3.5) and 4x on Lumina. Most notably, MAMBO-G accelerates the 14B-parameter Wan2.1 video model by 2x while preserving visual fidelity, offering a practical solution for efficient large-scale video synthesis. Our implementation follows a mainstream open-source diffusion framework and is plug-and-play with existing pipelines.

cs.CV↗

arXiv2026

Consistent inverse optimal control for infinite time-horizon discounted nonlinear systems under noisy observations

Inverse optimal control (IOC) aims to estimate the underlying cost that governs the observed behavior of an expert system. However, in practical scenarios, the collected data is often corrupted by noise, which poses significant challenges for accurate cost function recovery. In this work, we propose an IOC framework that effectively addresses the presence of observation noise. In particular, compared to our previous work \cite{wang2025consistent}, we consider the case of discrete-time, infinite-horizon, discounted MDPs whose transition kernel is only weak Feller. By leveraging the occupation measure framework, we first establish the necessary and sufficient optimality conditions for the expert policy and then construct an infinite dimensional optimization problem based on these conditions. This problem is then approximated by polynomials to get a finite-dimensional numerically solvable one, which relies on the moments of the state-action trajectory's occupation measure. More specifically, the moments are robustly estimated from the noisy observations by a combined misspecified Generalized Method of Moments (GMM) estimator derived from observation model and system dynamics. Consequently, the entire algorithm is based on convex optimization which alleviates the issues that arise from local minima and is asymptotically and statistically consistent. Finally, the performance of the proposed method is illustrated through numerical examples.

math.OC↗

arXiv2026

GRAB: An LLM-Inspired Sequence-First Click-Through Rate Prediction Modeling Paradigm

Traditional Deep Learning Recommendation Models (DLRMs) face increasing bottlenecks in performance and efficiency, often struggling with generalization and long-sequence modeling. Inspired by the scaling success of Large Language Models (LLMs), we propose Generative Ranking for Ads at Baidu (GRAB), an end-to-end generative framework for Click-Through Rate (CTR) prediction. GRAB integrates a novel Causal Action-aware Multi-channel Attention (CamA) mechanism to effectively capture temporal dynamics and specific action signals within user behavior sequences. Full-scale online deployment demonstrates that GRAB significantly outperforms established DLRMs, delivering a 3.05% increase in revenue and a 3.49% rise in CTR. Furthermore, the model demonstrates desirable scaling behavior: its expressive power shows a monotonic and approximately linear improvement as longer interaction sequences are utilized.

cs.IR↗

arXiv2026

PromptRL: Prompt Matters in RL for Flow-Based Image Generation

Flow matching models (FMs) have revolutionized text-to-image (T2I) generation, with reinforcement learning (RL) serving as a critical post-training strategy for alignment with reward objectives. In this research, we show that current RL pipelines for FMs suffer from two underappreciated yet important limitations: sample inefficiency due to insufficient generation diversity, and pronounced prompt overfitting, where models memorize specific training formulations and exhibit dramatic performance collapse when evaluated on semantically equivalent but stylistically varied prompts. We present PromptRL (Prompt Matters in RL for Flow-Based Image Generation), a framework that incorporates language models (LMs) as trainable prompt refinement agents directly within the flow-based RL optimization loop. This design yields two complementary benefits: rapid development of sophisticated prompt rewriting capabilities and, critically, a synergistic training regime that reshapes the optimization dynamics. PromptRL achieves state-of-the-art performance across multiple benchmarks, obtaining scores of 0.97 on GenEval, 0.98 on OCR accuracy, and 24.05 on PickScore. Furthermore, we validate the effectiveness of our RL approach on large-scale image editing models, improving the EditReward of FLUX.1-Kontext from 1.19 to 1.43 with only 0.06 million rollouts, surpassing Gemini 2.5 Flash Image (also known as Nano Banana), which scores 1.37, and achieving comparable performance with ReasonNet (1.44), which relied on fine-grained data annotations along with a complex multi-stage training. Our extensive experiments empirically demonstrate that PromptRL consistently achieves higher performance ceilings while requiring over 2$\times$ fewer rollouts compared to naive flow-only RL. Our code is available at https://github.com/G-U-N/UniRL.

cs.CV↗

arXiv2026

GEPO: Group Expectation Policy Optimization for Stable Heterogeneous Reinforcement Learning

As single-center computing approaches power constraints, decentralized training becomes essential. However, traditional Reinforcement Learning (RL) methods, crucial for enhancing large model post-training, cannot adapt to decentralized distributed training due to the tight coupling between parameter learning and rollout sampling. For this, we propose HeteroRL, a heterogeneous RL architecture that decouples these processes, enabling stable training across geographically distributed nodes connected via the Internet. The core component is Group Expectation Policy Optimization (GEPO), an asynchronous RL algorithm robust to latency caused by network delays or heterogeneity in computational resources. Our study reveals that high latency significantly increases KL divergence, leading to higher variance of importance weights and training instability. GEPO mitigates this issue by using group expectation weighting to exponentially reduce the variance of importance weights, with theoretical guarantees. Experiments show GEPO achieves superior stability - only a 3% performance drop from online to 1800s latency-and reduces the best-to-last gap by 85% versus GSPO (1.8 vs. 12.0) while attaining the highest scores, highlighting its effectiveness in decentralized, resource-heterogeneous environments.

cs.LG↗

arXiv2026

Unifying Perception and Action: A Hybrid-Modality Pipeline with Implicit Visual Chain-of-Thought for Robotic Action Generation

Vision-Language-Action (VLA) models built upon Chain-of-Thought (CoT) have achieved remarkable success in advancing general-purpose robotic agents, owing to its significant perceptual comprehension. Recently, since text-only CoT struggles to adequately capture scene details in complex spatial environments, a highly promising strategy involves leveraging visual priors to guide robotic action generation. Nevertheless, these strategies face two inherent challenges: (i) a modality gap between visual observations and low-level actions, and (ii) unstable training due to competing objectives between visual prediction and action generation. To address these challenges, we propose a Vision-Integrated Trajectory Alignment (VITA) framework that learns a shared discrete latent space for vision and action, enabling joint modeling of perception and motor control. VITA introduces a implicit visual CoT: autoregressively generated tokens is simultaneously decoded into future frames predictions and robot actions, thereby internalizing visual dynamics as an inductive bias for motion planning. Extensive experiments on simulated and real-world environments demonstrate state-of-the-art performance. VITA improves 14.5\%, 9.6\% and 12.1\% over existing baselines on CALVIN, LIBERO and SimplerEnv. Furthermore, VITA attains an average success rate of 80.5\% across six real-world tasks, demonstrating its potential as a generalist robotic manipulation model.

cs.RO↗

arXiv2026

Testing Non-Standard Neutrinos in Purely Leptonic Lepton Decays

We propose a method to probe sterile neutrinos using polarization observables in the purely leptonic decays $\ell^{\prime-} \to \ell^{-} \barν_{\ell} ν_{\ell^{\prime}}$. By analyzing angular distributions and asymmetries derived from polarized decay rates, we identify distinctive signatures of sterile neutrino mixing. In particular, we demonstrate that sterile neutrinos can induce singularities in certain asymmetry parameters as functions of the invariant mass squared of the neutrino pair. These singularities occur for sterile neutrino masses $m_{4ν}$ satisfying $m_{4ν}^2 < m_{\ell^{\prime}}^2 / 2$, providing a clear target for experimental investigation. Our results motivate the incorporation of polarized beam sources at future colliders to enhance sensitivity to sterile neutrinos and other new physics.

hep-ph↗

arXiv2026

MetaVLA: Unified Meta Co-training For Efficient Embodied Adaption

Vision-Language-Action (VLA) models show promise in embodied reasoning, yet remain far from true generalists-they often require task-specific fine-tuning, incur high compute costs, and generalize poorly to unseen tasks. We propose MetaVLA, a unified, backbone-agnostic post-training framework for efficient and scalable alignment. MetaVLA introduces Context-Aware Meta Co-Training, which consolidates diverse target tasks into a single fine-tuning stage while leveraging structurally diverse auxiliary tasks to improve in-domain generalization. Unlike naive multi-task SFT, MetaVLA integrates a lightweight meta-learning mechanism-derived from Attentive Neural Processes-to enable rapid adaptation from diverse contexts with minimal architectural change or inference overhead. On the LIBERO benchmark, MetaVLA with six auxiliary tasks outperforms OpenVLA by up to 8.0% on long-horizon tasks, reduces training steps from 240K to 75K, and cuts GPU time by ~76%. These results show that scalable, low-resource post-training is achievable-paving the way toward general-purpose embodied agents. Code will be available.

cs.AI↗

arXiv2026

Masked Depth Modeling for Spatial Perception

Spatial visual perception is a fundamental requirement in physical-world applications like autonomous driving and robotic manipulation, driven by the need to interact with 3D environments. Capturing pixel-aligned metric depth using RGB-D cameras would be the most viable way, yet it usually faces obstacles posed by hardware limitations and challenging imaging conditions, especially in the presence of specular or texture-less surfaces. In this work, we argue that the inaccuracies from depth sensors can be viewed as "masked" signals that inherently reflect underlying geometric ambiguities. Building on this motivation, we present LingBot-Depth, a depth completion model which leverages visual context to refine depth maps through masked depth modeling and incorporates an automated data curation pipeline for scalable training. It is encouraging to see that our model outperforms top-tier RGB-D cameras in terms of both depth precision and pixel coverage. Experimental results on a range of downstream tasks further suggest that LingBot-Depth offers an aligned latent representation across RGB and depth modalities. We release the code, checkpoint, and 3M RGB-depth pairs (including 2M real data and 1M simulated data) to the community of spatial perception.

cs.CV↗

arXiv2026

Exceptional-point-like Sensing near Hermitian Critical Points

A non-Hermitian system at an exceptional point (EP), a specific critical point (CP) associated with the parity-time symmetric phase transition, exhibits a sublinear response to perturbation and promise unprecedented sensitivity beyond the linear-response Hermitian sensors, so far operating at the diabolic points (DP). Despite great advancements, its sensitivity enhancement is fundamentally limited by the divergent Petermann factor, intrinsically rooted in the non-Hermitian eigenvector degeneracy, and practically by the system complexity. Here, we report the CP-resulting square-root response to the refractive index change and enhanced sensitivity in a simple chiral Hermitian cavity without phase transitions. Because of the inherent eigenvector orthogonality, this CP-based Hermitian sensor exhibits an EP-like response and enhanced sensitivity, breaking the Petermann-factor limit of sensitivity in non-Hermitian counterparts. This work paves the way towards exploring the Hermitian CPs for ultrasensitive sensing outperforming both the EP- and DP-based sensors.

physics.optics↗

arXiv2026

Trustworthy Longitudinal Brain MRI Completion: A Deformation-Based Approach with KAN-Enhanced Diffusion Model

Longitudinal brain MRI is essential for lifespan study, yet high attrition rates often lead to missing data, complicating analysis. Deep generative models have been explored, but most rely solely on image intensity, leading to two key limitations: 1) the fidelity or trustworthiness of the generated brain images are limited, making downstream studies questionable; 2) the usage flexibility is restricted due to fixed guidance rooted in the model structure, restricting full ability to versatile application scenarios. To address these challenges, we introduce DF-DiffCom, a Kolmogorov-Arnold Networks (KAN)-enhanced diffusion model that smartly leverages deformation fields for trustworthy longitudinal brain image completion. Trained on OASIS-3, DF-DiffCom outperforms state-of-the-art methods, improving PSNR by 5.6% and SSIM by 0.12. More importantly, its modality-agnostic nature allows smooth extension to varied MRI modalities, even to attribute maps such as brain tissue segmentation results.

cs.CV↗