arXiv subjects

Ke Zhang

Publications and source records attributed to Ke Zhang.

At least 91 records · Page 5Linked to original sources

arXiv2026

Rollout-Training Co-Design for Efficient LLM-Based Multi-Agent Reinforcement Learning

Despite algorithm-level innovations for multi-agent reinforcement learning (MARL), the underlying networked infrastructure for large-scale MARL training remains underexplored. Existing training frameworks primarily optimize for single-agent scenarios and fail to address the unique system-level challenges of MARL, including rollout-training synchronization barriers, rollout load imbalance, and training resource underutilization. To bridge this gap, we propose FlexMARL, the first end-to-end training framework that holistically optimizes rollout, training, and their orchestration for large-scale LLM-based MARL. Specifically, FlexMARL introduces the joint orchestrator to manage data flow under the rollout-training disaggregated architecture. Building upon the experience store, a novel micro-batch driven asynchronous pipeline eliminates the synchronization barriers while providing strong consistency guarantees. Rollout engine adopts a parallel sampling scheme combined with hierarchical load balancing, which adapts to skewed inter/intra-agent request patterns. Training engine achieves on-demand hardware binding through agent-centric resource allocation. The training states of different agents are swapped via unified and location-agnostic communication. Empirical results on a large-scale production cluster demonstrate that FlexMARL achieves up to 7.3x speedup and improves hardware utilization by up to 5.6x compared to existing frameworks.

cs.LG↗

arXiv2026

DexFormer: Cross-Embodied Dexterous Manipulation via History-Conditioned Transformer

Dexterous manipulation remains one of the most challenging problems in robotics, requiring coherent control of high-DoF hands and arms under complex, contact-rich dynamics. A major barrier is embodiment variability: different dexterous hands exhibit distinct kinematics and dynamics, forcing prior methods to train separate policies or rely on shared action spaces with per-embodiment decoder heads. We present DexFormer, an end-to-end, dynamics-aware cross-embodiment policy built on a modified transformer backbone that conditions on historical observations. By using temporal context to infer morphology and dynamics on the fly, DexFormer adapts to diverse hand configurations and produces embodiment-appropriate control actions. Trained over a variety of procedurally generated dexterous-hand assets, DexFormer acquires a generalizable manipulation prior and exhibits strong zero-shot transfer to Leap Hand, Allegro Hand, and Rapid Hand. Our results show that a single policy can generalize across heterogeneous hand embodiments, establishing a scalable foundation for cross-embodiment dexterous manipulation. Project website: https://davidlxu.github.io/DexFormer-web/.

cs.RO↗

arXiv2026

Dolphin-v2: Universal Document Parsing via Scalable Anchor Prompting

Document parsing has garnered widespread attention as vision-language models (VLMs) advance OCR capabilities. However, the field remains fragmented across dozens of specialized models with varying strengths, forcing users to navigate complex model selection and limiting system scalability. Moreover, existing two-stage approaches depend on axis-aligned bounding boxes for layout detection, failing to handle distorted or photographed documents effectively. To this end, we present Dolphin-v2, a two-stage document image parsing model that substantially improves upon the original Dolphin. In the first stage, Dolphin-v2 jointly performs document type classification (digital-born versus photographed) alongside layout analysis. For digital-born documents, it conducts finer-grained element detection with reading order prediction. In the second stage, we employ a hybrid parsing strategy: photographed documents are parsed holistically as complete pages to handle geometric distortions, while digital-born documents undergo element-wise parallel parsing guided by the detected layout anchors, enabling efficient content extraction. Compared with the original Dolphin, Dolphin-v2 introduces several crucial enhancements: (1) robust parsing of photographed documents via holistic page-level understanding, (2) finer-grained element detection (21 categories) with semantic attribute extraction such as author information and document metadata, and (3) code block recognition with indentation preservation, which existing systems typically lack. Comprehensive evaluations are conducted on DocPTBench, OmniDocBench, and our self-constructed RealDoc-160 benchmark. The results demonstrate substantial improvements: +14.78 points overall on the challenging OmniDocBench and 91% error reduction on photographed documents, while maintaining efficient inference through parallel processing.

cs.CV↗

arXiv2026

BAP-SRL: Bayesian Adaptive Priority Safe Reinforcement Learning for Vehicle Motion Planning at Mixed Traffic Intersections

Navigating urban intersections, especially when interacting with heterogeneous traffic participants, presents a formidable challenge for autonomous vehicles (AVs). In such environments, safety risks arise simultaneously from multiple sources, each carrying distinct priority levels and sensitivities that necessitate differential protection preferences. While safe reinforcement learning (RL) offers a robust paradigm for constrained decision-making, existing methods typically model safety as a single constraint or employ static, heuristic weighting schemes for multiple constraints. These approaches often fail to address the dynamic nature of multi-source risks, leading to gradient cancellation that hampers learning, and suboptimal trade-offs in critical dilemma zones. To address this, we propose a Bayesian adaptive priority safe reinforcement learning (BAP-SRL) based motion planning framework. Unlike heuristic weighting schemes, BAP formulates constraint prioritization as a probabilistic inference task. By modeling historical optimization difficulty as a Bayesian prior and instantaneous risk evidence as a likelihood, BAP dynamically gates gradient updates using a Bayesian inference mechanism on latent constraint criticality. Extensive experiments demonstrate that our approach outperforms state-of-the-art baselines in handling interactions with stochastic, heterogeneous agents, achieving lower collision rates and smoother conflict resolution.

eess.SY↗

arXiv2026

H$_2^{18}$O in the terrestrial planet-forming regions of protoplanetary disks

Isotopologues play an important role in solar system cosmochemistry studies, revealing details of early planet formation physics and chemistry. Oxygen isotopes, as measured in solar system materials, reveal evidence for both mass-dependent fractionation processes and a mass-independent process commonly attributed to isotope-selective photodissociation of CO in the solar nebula. The sensitivity of JWST's MIRI-MRS enables studies of isotopologues in the terrestrial planet-forming regions around nearby young stars. We report here on a search for H$_2^{18}$O in 22 disks from the JDISC Survey with evidence for substantial water vapor reservoirs, with the goal of measuring H$_2^{16}$O/H$_2^{18}$O ratios, and potentially revealing the predicted enhancement of H$_2^{18}$O caused by isotope-selective photodissociation. We find marginal detections of H$_2^{18}$O in six disks, and a more significant detection of H$_2^{18}$O in the disk around WSB 52. Modeling of the detected H$_2^{18}$O lines assuming an ISM ratio of H$_2^{16}$O/H$_2^{18}$O predicts H$_2^{18}$O features consistent with observations for four of the modeled disks, but stronger H$_2^{18}$O features than are observed in three of the modeled disks, which includes WSB 52. Therefore, these latter three disks require a higher H$_2^{16}$O/H$_2^{18}$O ratio than the ISM in the water-emitting region, in contrast to long-standing theoretical expectations. We suggest that either the H$_2^{18}$O-rich water has been removed from the emitting region and replaced by H$_2^{18}$O-poor water formed by reactions with $^{18}$O-poor CO, or that the gas-phase water is depleted in $^{18}$O via mass-dependent fractionation processes at the water snowline.

astro-ph.EP↗

arXiv2026

A Segmentation-driven Editing Method for Bolt Defect Augmentation and Detection

Bolt defect detection is critical to ensure the safety of transmission lines. However, the scarcity of defect images and imbalanced data distributions significantly limit detection performance. To address this problem, we propose a segmentationdriven bolt defect editing method (SBDE) to augment the dataset. First, a bolt attribute segmentation model (Bolt-SAM) is proposed, which enhances the segmentation of complex bolt attributes through the CLAHE-FFT Adapter (CFA) and Multipart- Aware Mask Decoder (MAMD), generating high-quality masks for subsequent editing tasks. Second, a mask optimization module (MOD) is designed and integrated with the image inpainting model (LaMa) to construct the bolt defect attribute editing model (MOD-LaMa), which converts normal bolts into defective ones through attribute editing. Finally, an editing recovery augmentation (ERA) strategy is proposed to recover and put the edited defect bolts back into the original inspection scenes and expand the defect detection dataset. We constructed multiple bolt datasets and conducted extensive experiments. Experimental results demonstrate that the bolt defect images generated by SBDE significantly outperform state-of-the-art image editing models, and effectively improve the performance of bolt defect detection, which fully verifies the effectiveness and application potential of the proposed method. The code of the project is available at https://github.com/Jay-xyj/SBDE.

cs.CV↗

arXiv2026

Clustering-Based User Selection in Federated Learning: Metadata Exploitation for 3GPP Networks

Federated learning (FL) enables collaborative model training without sharing raw user data, but conventional simulations often rely on unrealistic data partitioning and current user selection methods ignore data correlation among users. To address these challenges, this paper proposes a metadatadriven FL framework. We first introduce a novel data partition model based on a homogeneous Poisson point process (HPPP), capturing both heterogeneity in data quantity and natural overlap among user datasets. Building on this model, we develop a clustering-based user selection strategy that leverages metadata, such as user location, to reduce data correlation and enhance label diversity across training rounds. Extensive experiments on FMNIST and CIFAR-10 demonstrate that the proposed framework improves model performance, stability, and convergence in non-IID scenarios, while maintaining comparable performance under IID settings. Furthermore, the method shows pronounced advantages when the number of selected users per round is small. These findings highlight the framework's potential for enhancing FL performance in realistic deployments and guiding future standardization.

eess.SP↗

arXiv2026

Fast Collaborative Inference via Distributed Speculative Decoding

Speculative decoding accelerates large language model (LLM) inference by allowing a small draft model to predict multiple future tokens for verification by a larger target model. In AI-native radio access networks (AI-RAN), this enables device-edge collaborative inference but introduces significant uplink overhead, as existing distributed speculative decoding schemes transmit full vocabulary logits at every step. We propose a sparsify-then-sample strategy, Truncated Sparse Logits Transmission (TSLT), which transmits only the logits and indices of a truncated candidate set. We provide theoretical guarantees showing that the acceptance rate is preserved under TSLT. TSLT is further extended to multi-candidate case, where multiple draft candidates per step increase acceptance probability. Experiments show that TSLT significantly reduces uplink communication while maintaining end-to-end inference latency and model quality, demonstrating its effectiveness for scalable, communication-efficient distributed LLM inference in future AI-RAN systems.

eess.SP↗

arXiv2026

Multi-Agent Collaborative Reward Design for Enhancing Reasoning in Reinforcement Learning

We present CRM (Multi-Agent Collaborative Reward Model), a framework that replaces a single black-box reward model with a coordinated team of specialist evaluators to improve robustness and interpretability in RLHF. Conventional reward models struggle to jointly optimize multiple, sometimes conflicting, preference dimensions (e.g., factuality, helpfulness, safety) and offer limited transparency into why a score is assigned. CRM addresses these issues by decomposing preference evaluation into domain-specific agents that each produce partial signals, alongside global evaluators such as ranker-based and embedding-similarity rewards. A centralized aggregator fuses these signals at each timestep, balancing factors like step-wise correctness, multi-agent agreement, and repetition penalties, yielding a single training reward compatible with standard RL pipelines. The policy is optimized with advantage-based updates (e.g., GAE), while a value model regresses to the aggregated reward, enabling multi-perspective reward shaping without requiring additional human annotations beyond those used to train the evaluators. To support training and assessment, we introduce rewardBench, a benchmark and training suite aligned with the collaborative structure of CRM. Together, CRM and rewardBench provide a practical, modular path to more transparent reward modeling and more stable optimization.

cs.AI↗

arXiv2026

Protoplanetary disk cavities with JWST-MIRI: a dichotomy in molecular emission

The evolution of planet-forming regions in protoplanetary disks is of fundamental importance to understanding planet formation. Disks with a central deficit in dust emission, a "cavity", have long attracted interest as potential evidence for advanced disk clearing by protoplanets and/or winds. Before JWST, infrared spectra showed that these disks typically lack the strong molecular emission observed in full disks. In this work, we combine a sample of 12 disks with millimeter cavities of a range of sizes ($\sim2$-70 au) and different levels of millimeter and infrared continuum deficits. We analyze their molecular spectra as observed with MIRI on JWST, homogeneously reduced with the new JDISCS pipeline. This analysis demonstrates a stark dichotomy in molecular emission where "molecule-rich" (MR) cavities follow global trends between water, CO, and OH luminosity and accretion luminosity as in full disks, while "molecule-poor" (MP) cavities are significantly sub-luminous in all molecules except sometimes OH. Disk cavities generally show sub-luminous organic emission, higher OH/H$_2$O ratios, and suggest a lower water column density. The sub-thermal excitation of CO and water vibrational lines suggests a decreased gas density in the emitting layer in all cavities, supporting model expectations for C$_2$H$_2$ photodissociation. We discover a bifurcation in infrared index (lower in MR cavities) suggesting that the molecular dichotomy is linked to residual $μ$m-size dust within millimeter disk cavities. Put together, these results suggest a feedback process between dust depletion, gas density decrease, and molecule dissociation. Disk cavities may have a common evolutionary sequence where MR switch into MP over time.

astro-ph.EP↗

arXiv2026

Tracing Pebble Drift History in Two Protoplanetary Disks with CO Enhancement

Pebble drift is an important mechanism for supplying the materials needed to build planets in the inner region of protoplanetary disks. Thus, constraining pebble drift's timescales and mass flux is essential to understanding planet formation history. Current pebble drift models suggest pebble fluxes can be constrained from the enhancement of gaseous volatile abundances when icy pebbles sublimate after drifting across key snowlines. In this work, we present ALMA observations of spatially resolved $^{13}$C$^{18}$O J=2-1 line emission inside the midplane CO snowline of the HD 163296 and MWC 480 protoplanetary disks. We use radiative transfer and thermochemical models to constrain the spatial distribution of CO gas column density. We find that both disks display centrally peaked CO abundance enhancement of up to ten times of ISM abundance levels. For HD 163296 and MWC 480, the inferred enhancements require 250-350 and 480-660 Earth Masses of pebbles to have drifted across their CO snowlines, respectively. These ranges fall within cumulative pebble mass flux ranges to grow gas giants in the interior to the CO snowline. The centrally peaked CO enhancement is unexpected in current pebble drift models, which predict CO enhancement peaks at the CO snowline or is uniform inside the snowline. We propose two hypotheses to explain the centrally-peaked CO enhancement, including a large CO desorption distance and CO trapped in water ice. By testing both hypotheses with the 1D gas and dust evolution code chemcomp, we find that volatile trapping (about 30\%) best reproduces the centrally peaked CO enhancement observed.

astro-ph.EP↗

arXiv2025

Multi-Component Open/Relative/Local Correspondence

For a toric Calabi-Yau 3-orbifold relative to s Aganagic-Vafa outer branes, we prove a correspondence among the genus-zero open Gromov-Witten invariants with maximal winding at each brane and: (i) closed invariants of a toric Calabi-Yau (3+s)-orbifold; (ii) formal relative invariants of a formal toric Calabi-Yau (FTCY) 3-orbifold with maximal tangency to s divisors; (iii) formal relative invariants of a sequence of FTCY intermediate geometries interpolating dimensions 3 and 3+s. The correspondence provides examples of the log/local principle of van Garrel-Graber-Ruddat in the multi-component setting and the refined conjecture of Brini-Bousseau-van Garrel via intermediate geometries. It also establishes the multi-component case of the open/closed correspondence proposed by Lerche-Mayr and studied by Liu-Yu. As an application, we obtain examples of the conjecture of Klemm-Pandharipande on the integrality of BPS invariants of higher-dimensional toric Calabi-Yau manifolds. Along the way, we set the basic stages of the relative Gromov-Witten theory of higher-dimensional FTCY orbifolds, generalizing the case of smooth 3-folds by Li-Liu-Liu-Zhou.

math.AG↗

arXiv2025

Endless World: Real-Time 3D-Aware Long Video Generation

Producing long, coherent video sequences with stable 3D structure remains a major challenge, particularly in streaming scenarios. Motivated by this, we introduce Endless World, a real-time framework for infinite, 3D-consistent video generation.To support infinite video generation, we introduce a conditional autoregressive training strategy that aligns newly generated content with existing video frames. This design preserves long-range dependencies while remaining computationally efficient, enabling real-time inference on a single GPU without additional training overhead.Moreover, our Endless World integrates global 3D-aware attention to provide continuous geometric guidance across time. Our 3D injection mechanism enforces physical plausibility and geometric consistency throughout extended sequences, addressing key challenges in long-horizon and dynamic scene synthesis.Extensive experiments demonstrate that Endless World produces long, stable, and visually coherent videos, achieving competitive or superior performance to existing methods in both visual fidelity and spatial consistency. Our project has been available on https://bwgzk-keke.github.io/EndlessWorld/.

cs.CV↗

arXiv2025

CDKFormer: Contextual Deviation Knowledge-Based Transformer for Long-Tail Trajectory Prediction

Predicting the future movements of surrounding vehicles is essential for ensuring the safe operation and efficient navigation of autonomous vehicles (AVs) in urban traffic environments. Existing vehicle trajectory prediction methods primarily focus on improving overall performance, yet they struggle to address long-tail scenarios effectively. This limitation often leads to poor predictions in rare cases, significantly increasing the risk of safety incidents. Taking Argoverse 2 motion forecasting dataset as an example, we first investigate the long-tail characteristics in trajectory samples from two perspectives, individual motion and group interaction, and deriving deviation features to distinguish abnormal from regular scenarios. On this basis, we propose CDKFormer, a Contextual Deviation Knowledge-based Transformer model for long-tail trajectory prediction. CDKFormer integrates an attention-based scene context fusion module to encode spatiotemporal interaction and road topology. An additional deviation feature fusion module is proposed to capture the dynamic deviations in the target vehicle status. We further introduce a dual query-based decoder, supported by a multi-stream decoder block, to sequentially decode heterogeneous scene deviation features and generate multimodal trajectory predictions. Extensive experiments demonstrate that CDKFormer achieves state-of-the-art performance, significantly enhancing prediction accuracy and robustness for long-tailed trajectories compared to existing methods, thus advancing the reliability of AVs in complex real-world environments.

cs.RO↗

arXiv2025

Tool-RoCo: An Agent-as-Tool Self-organization Large Language Model Benchmark in Multi-robot Cooperation

This study proposes Tool-RoCo, a novel benchmark for evaluating large language models (LLMs) in long-term multi-agent cooperation based on RoCo, a multi-robot cooperative benchmark. Recent research on LLM-based multi-agent systems has relied on predefined orchestration, while ignoring agent autonomy. Tool-RoCo treats other agents as tools and introduces cooperative tools, leveraging tool usage to evaluate multi-agent cooperation and self-organization. Tool usage means that each agent (LLM) selects a tool from a candidate set based on the current state, receives feedback, and adjusts its selection in subsequent rounds. To evaluate different autonomy levels, we propose four LLM paradigms: (1) centralized cooperation, where a single LLM allocates tools to all agents; (2) centralized self-organization, where a central LLM autonomously activates agents while keeping others inactive; (3) decentralized cooperation, where each agent has its own LLM and calls tools based on local information; and (4) self-organization, where a randomly chosen initial agent can request collaboration, activating additional agents via tool calls. Tool-RoCo includes three multi-robot tasks, SORT, PACK, and CABINET, to measure format and parameter accuracy and agent coordination through tool usage. The results using several LLMs showed that cooperative tools accounted for only 7.09% of all tools, indicating that LLM-based agents rarely invoked others as assistants. Moreover, activation tools accounted for 96.42%, suggesting that current LLMs tend to maintain active agents while seldom deactivating them for adaptive coordination. Tool-RoCo provides a systematic benchmark to evaluate LLM autonomy and cooperation in multi-agent tasks. Code and Demo: https://github.com/ColaZhang22/Tool-Roco

cs.MA↗

arXiv2025

OOCO: Latency-disaggregated Architecture for Online-Offline Co-locate LLM Serving

Large Language Models (LLMs) are increasingly deployed in both latency-sensitive online services and cost-sensitive offline workloads. Co-locating these workloads on shared serving instances can improve resource utilization, but directly applying this approach to Prefill/Decode (P/D) disaggregated systems introduces severe load imbalance, as fluctuating request mixes alter the intrinsic P/D ratio. Existing dynamic adjustment techniques cannot keep up with the bursty traffic patterns of online services. We propose a latency-constraint disaggregated architecture, which separates cluster resources into latency-strict and latency-relaxed pools based on task latency requirements. This design enables flexible placement of offline decode tasks, mitigating P/D imbalance while preserving online performance. To fully exploit this flexibility, we propose (1) a bottleneck-based scheduler guided by a Roofline-based performance model for performance bottleneck based scheduling, and (2) a fast preemption mechanism that strictly enforces Service Level Objectives (SLOs) for online requests. Experiments on real-world traces show that compared to existing offline system approaches, our method improves offline throughput by up to 3x, while maintaining online request SLOs.

cs.DC↗

arXiv2025

A Survey on Speech Large Language Models for Understanding

Speech understanding is essential for interpreting the diverse forms of information embedded in spoken language, including linguistic, paralinguistic, and non-linguistic cues that are vital for effective human-computer interaction. The rapid advancement of large language models (LLMs) has catalyzed the emergence of Speech Large Language Models (Speech LLMs), which marks a transformative shift toward general-purpose speech understanding systems. To further clarify and systematically delineate task objectives, in this paper, we formally define the concept of speech understanding and introduce a structured taxonomy encompassing its informational, functional, and format dimensions. Within this scope of definition, we present a comprehensive review of current Speech LLMs, analyzing their architectures through a three-stage abstraction: Modality Feature Extraction, Modality Information Fusion, and LLM Inference. In addition, we examine training strategies, discuss representative datasets, and review evaluation methodologies adopted in the field. Based on empirical analyses and experimental evidence, we identify two key challenges currently facing Speech LLMs: instruction sensitivity and degradation in semantic reasoning and propose concrete directions for addressing these issues. Through this systematic and detailed survey, we aim to offer a foundational reference for researchers and practitioners working toward more robust, generalizable, and human-aligned Speech LLMs.

eess.AS↗

arXiv2025

A Submillimeter Survey of CS Excitation in Protoplanetary Disks: Evidence of X-ray-Driven Sulfur Chemistry

The sulfur chemistry in protoplanetary disks influences the properties of nascent planets, including potential habitability. Although the inventory of sulfur molecules in disks has gradually increased over the last decade, CS is still the most commonly-observed sulfur-bearing species and it is expected to be the dominant gas-phase sulfur carrier beyond the water snowline. Despite this, few dedicated multi-line observations exist, and thus the typical disk CS chemistry is not well constrained. Moreover, it is unclear how that chemistry - and in turn, the bulk volatile sulfur reservoir - varies with stellar and disk properties. Here, we present the largest survey of CS to date, combining both new and archival observations from ALMA, SMA, and NOEMA of 12 planet-forming disks, covering a range of stellar spectral types and dust morphologies. Using these data, we derived disk-integrated CS gas excitation conditions in each source. Overall, CS chemistry appears similar across our sample with rotational temperatures of ${\approx}$10-40 K and column densities between 10$^{12}$-10$^{13}$ cm$^{-2}$. CS column densities do not show strong trends with most source properties, which broadly suggests that CS chemistry is not highly sensitive to disk structure or stellar characteristics. We do, however, identify a positive correlation between stellar X-ray luminosity and CS column density, which indicates that the dominant CS formation pathway is likely via ion-neutral reactions in the upper disk layers, where X-ray-enhanced S$^+$ and C$^+$ drive abundant CS production. Thus, using CS as a tracer of gas-phase sulfur abundance requires a nuanced approach that accounts for its emitting region and dependence on X-ray luminosity.

astro-ph.EP↗