arXiv subjects

Yuheng Zhao

Publications and source records attributed to Yuheng Zhao.

At least 19 recordsLinked to original sources

arXiv2026

Insurance as AI Risk Infrastructure: A Generative-Agent Simulation of AI Adoption

The rapid evolution of artificial intelligence (AI) tools has demonstrated immense potential to enhance societal well-being and operational efficiency. However, the inherent unreliability and uncertain operational consequences of modern AI systems, typified by large language models (LLMs), have created a significant barrier to enterprise adoption. Many enterprises remain hesitant to integrate these tools deeply into their workflows due to concerns about unpredictable losses and liability exposure. While existing technical safeguards primarily seek to reduce the likelihood or severity of AI-enabled workflow failures, they do not by themselves provide ex post financial protection when residual pecuniary tail losses materialize. In this paper, we introduce a socio-economic framework that complements these safeguards by transferring and absorbing the residual financial consequences of AI adoption through insurance. To evaluate this framework, we develop an LLM-driven agent-based social simulation (LABSS) system. We assess the behavioral validity of the simulation using established economic and sociological theories. Our analysis demonstrates that the proposed insurance framework reduces firm-level financial exposure, thereby accelerating the aggregate adoption of AI tools and improving firm solvency and aggregate capital.

cs.MA↗

arXiv2026

Rollplex: Cross-Phase GPU Spatial Sharing for Vision Language Model Post-Training

Vision-language models (VLMs) enable embodied agents to reason and act from visual observations and language instructions. Reinforcement learning (RL) post-training enhances these capabilities using task feedback, but current on-policy RL runtimes execute rollout, reference scoring, and actor training in strict serial phases. While effective for text-only RL, this phase-granular execution is wasteful for VLMs, where processing dense video inputs and prompt prefixes occupies a large fraction of each phase. Because prefix processing is independent of the generated response, it can be run alongside rollout decoding, which leaves GPU compute capacity underutilized, without breaking synchronous on-policy semantics. We present Rollplex, a runtime that decomposes the reference and training phase and moves the prefix computation into the rollout decode window. Realizing this schedule requires more than concurrent kernel launches: naive colocation of Qwen2.5-VL-32\,B requires roughly 165\,GiB per GPU, while rollout and training prefer different tensor-parallel (TP) degrees and weight layouts. Rollplex addresses these constraints with two mechanisms. Phase-aware memory management controls HBM residency according to producer--consumer lifetimes. Parallelism-aware weight sharing uses the same physical storage for layout-compatible tensors across distinct TP degrees and reconstructs only incompatible tensors, avoiding a complete second actor copy. On 32 H800 GPUs, Rollplex achieves $1.23\times$--$1.30\times$ speedup over serial colocation and $1.57\times$--$2.24\times$ over disaggregation under the same GPU budget, while preserving the synchronous RL update.

cs.LG↗

arXiv2026

ROSE: Rollout On Serving GPUs via Cooperative Elasticity for Agentic RL

Agentic reinforcement learning (RL) is reshaping LLM post-training, but end-to-end training time is dominated by compute-intensive, multi-turn rollouts whose resource demand varies significantly across training steps. Resource-fixed systems cannot adapt to this variation, while resource-elastic approaches that provision external GPUs on demand suffer from high allocation overhead and limited availability. We observe that serving clusters leave substantial GPU compute and memory idle, and propose cooperative elasticity: sharing already-deployed serving GPUs with rollout workloads to provide on-demand elastic capacity. Realizing this is non-trivial, as it must preserve serving SLOs under bursty traffic while minimizing cross-cluster communication overhead. We present ROSE, a system that realizes cooperative elasticity for agentic RL post-training, comprising three components: (1) an SLO-safe co-serving executor that co-locates heterogeneous serving and rollout models on the same GPUs, dynamically sharing memory and compute while preserving serving SLOs; (2) a cross-cluster weight transfer engine that leverages shard-aware routing and weight sparsity for fast synchronization; and (3) an elastic rollout scheduler that dynamically routes rollouts across dedicated and opportunistic serving GPUs. Experiments across multiple model sizes and cluster scales show that ROSE improves end-to-end throughput by 1.3 - 3.3 x over resource-fixed baselines and reduces rollout time by 1.2 - 1.5 x over resource-elastic baselines, with no serving SLO violations.

cs.DC↗

arXiv2026

Towards Fully Parameter-Free Stochastic Optimization: Grid Search with Self-Bounding Analysis

Parameter-free stochastic optimization aims to design algorithms that are agnostic to the underlying problem parameters while still achieving convergence rates competitive with optimally tuned methods. While some parameter-free methods do not require the specific values of the problem parameters, they still rely on prior knowledge, such as the lower or upper bounds of them. We refer to such methods as ``partially parameter-free''. In this work, we target achieving ``fully parameter-free'' methods, i.e., the algorithmic inputs do not need to satisfy any unverifiable condition related to the true problem parameters. We propose a powerful and general grid search framework, named \textsc{Grasp}, with a novel self-bounding analysis technique that effectively determines the search ranges of parameters, in contrast to previous work. Our method demonstrates generality in: (i) the non-convex case, where we propose a fully parameter-free method that achieves near-optimal convergence rate, up to logarithmic factors; (ii) the convex case, where our parameter-free methods are competitive with strong performance in terms of acceleration and universality. Finally, we contribute a sharper guarantee for the model ensemble, a final step of the grid search framework, under interpolated variance characterization.

cs.LG↗

arXiv2026

Intelligent Drill-Down: Large Language Model-Driven Drill-Down Technique for Human-AI Collaborative Visual Exploration

In visual analytics, applying filters to drill-down and extract higher-value insights is a common and important data analysis method. When the drill-down space becomes excessively large, analysts may lose orientation, leading to decreased efficiency in the drill-down process. To tackle these challenges, we propose the Intelligent Drill-Down Framework, in which a large language model (LLM) facilitates the generation of visual insights, leverages user interaction data to interpret user intent, and generates appropriate drill-down paths. Our method is designed to assist users in identifying valuable drill-down paths when exploring multidimensional data, thereby reducing the cognitive burden of data interpretation and facilitating the generation of insights. Specifically, we propose a drill-down path recommendation method, in which the LLM is trained to approximate a validated greedy algorithm. Secondly, we analyze the user's intent to construct a drill-down chart. Finally, we design a branch management method. Building upon this framework, we designed a system that includes a hybrid interface providing hierarchical navigation to monitor users and manage parallel branches, a visualization panel for interactive data exploration, and an insight panel to present analytical findings and generate drill-down recommendations. We evaluated the effectiveness of our method through a demonstrative use case and a user study.

cs.HC↗

arXiv2026

Gradient-Variation Regret Bounds for Unconstrained Online Learning

We develop parameter-free algorithms for unconstrained online learning with regret guarantees that scale with the gradient variation $V_T(u) = \sum_{t=2}^T \|\nabla f_t(u)-\nabla f_{t-1}(u)\|^2$. For $L$-smooth convex losses, we provide fully-adaptive algorithms achieving regret of $\widetilde{O}(\|u\|\sqrt{V_T(u)} + L\|u\|^2+G^4)$ without requiring prior knowledge of comparator norm $\|u\|$, Lipschitz constant $G$, or smoothness $L$. The update in each round can be computed efficiently via a closed-form expression. Our results extend to dynamic regret and find immediate implications for the stochastically-extended adversarial (SEA) model, which significantly improves upon the previous best-known result (Wang et al., 2025).

cs.LG↗

arXiv2026

InfoAlign: A Human-AI Co-Creation System for Storytelling with Infographics

Storytelling infographics are a powerful medium for communicating data-driven stories through visual presentation. However, existing authoring tools lack support for maintaining story consistency and aligning with users' story goals throughout the design process. To address this gap, we conducted formative interviews and a quantitative analysis to identify design needs and common story-informed layout patterns in infographics. Based on these insights, we propose a narrative-centric workflow for infographic creation consisting of three phases: story construction, visual encoding, and spatial composition. Building on this workflow, we developed InfoAlign, a human-AI co-creation system that transforms long or unstructured text into stories, recommends semantically aligned visual designs, and generates layout blueprints. Users can intervene and refine the design at any stage, ensuring their intent is preserved and the infographic creation process remains transparent. Evaluations show that InfoAlign preserves story coherence across authoring stages and effectively supports human-AI co-creation for storytelling infographic design.

cs.HC↗

arXiv2025

RollArt: Disaggregated Multi-Task Agentic RL Training at Scale

Agentic Reinforcement Learning (RL) trains LLMs through multi-turn interactions with environments, producing workloads that mix compute-bound prefill, bandwidth-bound decoding, CPU-heavy environment execution, and bursty reward evaluation. Existing systems either colocate all stages on a single GPU cluster or decouple them only at a coarse granularity, overlooking hardware heterogeneity and incurring substantial synchronization overhead across stages. We present ROLLART, a system for multi-task agentic RL on disaggregated infrastructure. ROLLART maps each pipeline stage to best-fit hardware, routing prefill-heavy tasks to compute-optimized GPUs, decode-heavy tasks to bandwidth-optimized GPUs, and environments to CPU clusters. It decouples rollout at the trajectory level, allowing generation, environment interaction, and reward scoring to proceed independently, so that slow or failed environments never block the others. ROLLART offloads stateless reward computation to serverless infrastructure and overlaps rollout with training via staleness-bounded asynchronous weight synchronization. Our results demonstrate that ROLLART effectively improves training throughput and achieves 1.31--2.05 $\times$ training time reduction compared to various RL systems. We also evaluated ROLLART by training a hundreds-of-billions-parameter MoE model for Qoder product on an Alibaba cluster with above 3,000 GPUs, demonstrating its stability and scalability.

cs.DC↗

arXiv2025

OMP: One-step Meanflow Policy with Directional Alignment

Robot manipulation has increasingly adopted data-driven generative policy frameworks, yet the field faces a persistent trade-off: diffusion models suffer from high inference latency, while flow-based methods often require complex architectural constraints. Although in image generation domain, the MeanFlow paradigm offers a path to single-step inference, its direct application to robotics is impeded by critical theoretical pathologies, specifically spectral bias and gradient starvation in low-velocity regimes. To overcome these limitations, we propose the One-step MeanFlow Policy (OMP), a novel framework designed for high-fidelity, real-time manipulation. We introduce a lightweight directional alignment mechanism to explicitly synchronize predicted velocities with true mean velocities. Furthermore, we implement a Differential Derivation Equation (DDE) to approximate the Jacobian-Vector Product (JVP) operator, which decouples forward and backward passes to significantly reduce memory complexity. Extensive experiments on the Adroit and Meta-World benchmarks demonstrate that OMP outperforms state-of-the-art methods in success rate and trajectory accuracy, particularly in high-precision tasks, while retaining the efficiency of single-step generation.

cs.RO↗

arXiv2025

RollMux: Phase-Level Multiplexing for Disaggregated RL Post-Training

Rollout-training disaggregation is emerging as the standard architecture for Reinforcement Learning (RL) post-training, where memory-bound rollout and compute-bound training are physically disaggregated onto purpose-built clusters to maximize hardware efficiency. However, the strict synchronization required by on-policy algorithms introduces severe dependency bubbles, forcing one cluster to idle while the dependent phase is running on the other. We present RollMux, a cluster scheduling framework that reclaims these bubbles through cross-cluster orchestration. RollMux is built on the insight that the structural idleness of one job can be effectively utilized by the active phase of another. To realize this, we introduce the co-execution group abstraction, which partitions the cluster into isolated locality domains. This abstraction enables a two-tier scheduling architecture: an inter-group scheduler that optimizes job placement using conservative stochastic planning, and an intra-group scheduler that orchestrates a provably optimal round-robin schedule. The group abstraction also imposes a residency constraint, ensuring that massive model states remain cached in host memory to enable "warm-star" context switching. We evaluate RollMux on a production-scale testbed with 328 H20 and 328 H800 GPUs. RollMux improves cost efficiency by 1.84x over standard disaggregation and 1.38x over state-of-the-art co-located baselines, all while achieving 100% SLO attainment.

cs.DC↗

arXiv2025

Gradient-Variation Online Adaptivity for Accelerated Optimization with H\"older Smoothness

Smoothness is known to be crucial for acceleration in offline optimization, and for gradient-variation regret minimization in online learning. Interestingly, these two problems are actually closely connected -- accelerated optimization can be understood through the lens of gradient-variation online learning. In this paper, we investigate online learning with H\"older smooth functions, a general class encompassing both smooth and non-smooth (Lipschitz) functions, and explore its implications for offline optimization. For (strongly) convex online functions, we design the corresponding gradient-variation online learning algorithm whose regret smoothly interpolates between the optimal guarantees in smooth and non-smooth regimes. Notably, our algorithms do not require prior knowledge of the H\"older smoothness parameter, exhibiting strong adaptivity over existing methods. Through online-to-batch conversion, this gradient-variation online adaptivity yields an optimal universal method for stochastic convex optimization under H\"older smoothness. However, achieving universality in offline strongly convex optimization is more challenging. We address this by integrating online adaptivity with a detection-based guess-and-check procedure, which, for the first time, yields a universal offline method that achieves accelerated convergence in the smooth regime while maintaining near-optimal convergence in the non-smooth one.

cs.LG↗

arXiv2025

RollPacker: Mitigating Long-Tail Rollouts for Fast, Synchronous RL Post-Training

Reinforcement Learning (RL) is a pivotal post-training technique for enhancing the reasoning capabilities of Large Language Models (LLMs). However, synchronous RL post-training often suffers from significant GPU underutilization, referred to as bubbles, caused by imbalanced response lengths within rollout steps. Many RL systems attempt to alleviate this problem by relaxing synchronization, but this can compromise training accuracy. In this paper, we introduce tail batching, a novel rollout scheduling strategy for synchronous RL that systematically consolidates prompts leading to long-tail responses into a small subset of rollout steps (long rounds), while ensuring that the majority of steps (short rounds) involve only balanced, short rollouts. By excluding long responses from short rounds and rescheduling them into a few designated long rounds, tail batching effectively reduces GPU idle time during rollouts and significantly accelerates RL training without sacrificing accuracy. We present RollPacker, a system that fully harnesses the benefits of tail batching through holistic optimizations across all three RL stages: elastic parallelism adaptation for rollout, dynamic resource allocation and scheduling for reward, and stream-based training. Empirical results show that RollPacker achieves a 2.03x-2.56x end-to-end training time reduction compared to veRL and up to 2.24x speedup compared to RLHFuse for the Qwen2.5 family of LLMs on up to 128 H800 GPUs.

cs.DC↗

arXiv2025

KinemaFX: A Kinematic-Driven Interactive System for Particle Effect Exploration and Customization

Particle effects are widely used in games and animation to simulate natural phenomena or stylized visual effects. However, creating effect artworks is challenging for non-expert users due to their lack of specialized skills, particularly in finding particle effects with kinematic behaviors that match their intent. To address these issues, we present KinemaFX, a kinematic-driven interactive system, to assist non-expert users in constructing customized particle effect artworks. We propose a conceptual model of particle effects that captures both semantic features and kinematic behaviors. Based on the model, KinemaFX adopts a workflow powered by Large Language Models (LLMs) that supports intent expression through combined semantic and kinematic inputs, while enabling implicit preference-guided exploration and subsequent creation of customized particle effect artworks based on exploration results. Additionally, we developed a kinematic-driven method to facilitate efficient interactive particle effect search within KinemaFX via structured representation and measurement of particle effects. To evaluate KinemaFX, we illustrate usage scenarios and conduct a user study employing an ablation approach. Evaluation results demonstrate that KinemaFX effectively supports users in efficiently and customarily creating particle effect artworks.

cs.HC↗

arXiv2025

ProactiveVA: Proactive Visual Analytics with LLM-Based UI Agent

Visual analytics (VA) is typically applied to complex data, thus requiring complex tools. While visual analytics empowers analysts in data analysis, analysts may get lost in the complexity occasionally. This highlights the need for intelligent assistance mechanisms. However, even the latest LLM-assisted VA systems only provide help when explicitly requested by the user, making them insufficiently intelligent to offer suggestions when analysts need them the most. We propose a ProactiveVA framework in which LLM-powered UI agent monitors user interactions and delivers context-aware assistance proactively. To design effective proactive assistance, we first conducted a formative study analyzing help-seeking behaviors in user interaction logs, identifying when users need proactive help, what assistance they require, and how the agent should intervene. Based on this analysis, we distilled key design requirements in terms of intent recognition, solution generation, interpretability and controllability. Guided by these requirements, we develop a three-stage UI agent pipeline including perception, reasoning, and acting. The agent autonomously perceives users' needs from VA interaction logs, providing tailored suggestions and intuitive guidance through interactive exploration of the system. We implemented the framework in two representative types of VA systems, demonstrating its generalizability, and evaluated the effectiveness through an algorithm evaluation, case and expert study and a user study. We also discuss current design trade-offs of proactive VA and areas for further exploration.

cs.HC↗

arXiv2025

SceneLoom: Communicating Data with Scene Context

In data-driven storytelling contexts such as data journalism and data videos, data visualizations are often presented alongside real-world imagery to support narrative context. However, these visualizations and contextual images typically remain separated, limiting their combined narrative expressiveness and engagement. Achieving this is challenging due to the need for fine-grained alignment and creative ideation. To address this, we present SceneLoom, a Vision-Language Model (VLM)-powered system that facilitates the coordination of data visualization with real-world imagery based on narrative intents. Through a formative study, we investigated the design space of coordination relationships between data visualization and real-world scenes from the perspectives of visual alignment and semantic coherence. Guided by the derived design considerations, SceneLoom leverages VLMs to extract visual and semantic features from scene images and data visualization, and perform design mapping through a reasoning process that incorporates spatial organization, shape similarity, layout consistency, and semantic binding. The system generates a set of contextually expressive, image-driven design alternatives that achieve coherent alignments across visual, semantic, and data dimensions. Users can explore these alternatives, select preferred mappings, and further refine the design through interactive adjustments and animated transitions to support expressive data communication. A user study and an example gallery validate SceneLoom's effectiveness in inspiring creative design and facilitating design externalization.

cs.HC↗

arXiv2025

PCLVis: Visual Analytics of Process Communication Latency in Large-Scale Simulation

Large-scale simulations on supercomputers have become important tools for users. However, their scalability remains a problem due to the huge communication cost among parallel processes. Most of the existing communication latency analysis methods rely on the physical link layer information, which is only available to administrators. In this paper, a framework called PCLVis is proposed to help general users analyze process communication latency (PCL) events. Instead of the physical link layer information, the PCLVis uses the MPI process communication data for the analysis. First, a spatial PCL event locating method is developed. All processes with high correlation are classified into a single cluster by constructing a process-correlation tree. Second, the propagation path of PCL events is analyzed by constructing a communication-dependency-based directed acyclic graph (DAG), which can help users interactively explore a PCL event from the temporal evolution of a located PCL event cluster. In this graph, a sliding window algorithm is designed to generate the PCL events abstraction. Meanwhile, a new glyph called the communication state glyph (CS-Glyph) is designed for each process to show its communication states, including its in/out messages and load balance. Each leaf node can be further unfolded to view additional information. Third, a PCL event attribution strategy is formulated to help users optimize their simulations. The effectiveness of the PCLVis framework is demonstrated by analyzing the PCL events of several simulations running on the TH-1A supercomputer. By using the proposed framework, users can greatly improve the efficiency of their simulations.

cs.CV↗

arXiv2025

Reinforcement Learning Optimization for Large-Scale Learning: An Efficient and User-Friendly Scaling Library

We introduce ROLL, an efficient, scalable, and user-friendly library designed for Reinforcement Learning Optimization for Large-scale Learning. ROLL caters to three primary user groups: tech pioneers aiming for cost-effective, fault-tolerant large-scale training, developers requiring flexible control over training workflows, and researchers seeking agile experimentation. ROLL is built upon several key modules to serve these user groups effectively. First, a single-controller architecture combined with an abstraction of the parallel worker simplifies the development of the training pipeline. Second, the parallel strategy and data transfer modules enable efficient and scalable training. Third, the rollout scheduler offers fine-grained management of each sample's lifecycle during the rollout stage. Fourth, the environment worker and reward worker support rapid and flexible experimentation with agentic RL algorithms and reward designs. Finally, AutoDeviceMapping allows users to assign resources to different models flexibly across various stages.

cs.LG↗

arXiv2024

LightVA: Lightweight Visual Analytics with LLM Agent-Based Task Planning and Execution

Visual analytics (VA) requires analysts to iteratively propose analysis tasks based on observations and execute tasks by creating visualizations and interactive exploration to gain insights. This process demands skills in programming, data processing, and visualization tools, highlighting the need for a more intelligent, streamlined VA approach. Large language models (LLMs) have recently been developed as agents to handle various tasks with dynamic planning and tool-using capabilities, offering the potential to enhance the efficiency and versatility of VA. We propose LightVA, a lightweight VA framework that supports task decomposition, data analysis, and interactive exploration through human-agent collaboration. Our method is designed to help users progressively translate high-level analytical goals into low-level tasks, producing visualizations and deriving insights. Specifically, we introduce an LLM agent-based task planning and execution strategy, employing a recursive process involving a planner, executor, and controller. The planner is responsible for recommending and decomposing tasks, the executor handles task execution, including data analysis, visualization generation and multi-view composition, and the controller coordinates the interaction between the planner and executor. Building on the framework, we develop a system with a hybrid user interface that includes a task flow diagram for monitoring and managing the task planning process, a visualization panel for interactive data exploration, and a chat view for guiding the model through natural language instructions. We examine the effectiveness of our method through a usage scenario and an expert study.

cs.HC↗