Searcharxiv⌕ Search

arXiv subjects

Zheng Liu

Publications and source records attributed to Zheng Liu.

At least 55 records · Page 3Linked to original sources
arXiv2026

Machine Learning Guided Cooling System Optimization for Data Center

Effective data center cooling is crucial for reliable operation; however, cooling systems often exhibit inefficiencies that result in excessive energy consumption. This paper presents a three-stage, physics-guided machine learning framework for identifying and reducing cooling energy waste in high-performance computing facilities. Using one year of 10-minute resolution operational data from the Frontier exascale supercomputer, we first train a monotonicity-constrained gradient boosting surrogate that predicts facility accessory power from coolant flow rates, temperatures, and server power. The surrogate achieves a mean absolute error of 0.026 MW and predicts power usage effectiveness within 0.01 of measured values for 98.7% of test samples. In the second stage, the surrogate serves as a physics-consistent baseline to quantify excess cooling energy, revealing approximately 85 MWh of annual inefficiency concentrated in specific months, hours, and operating regimes. The third stage evaluates guardrail-constrained counterfactual adjustments to supply temperature and subloop flows, demonstrating that up to 96% of identified excess can be recovered through small, safe setpoint changes while respecting thermal limits and operational constraints. The framework yields interpretable recommendations, supports counterfactual analyses such as flow reduction during low-load periods and redistribution of thermal duty across cooling loops, and provides a practical pathway toward quantifiable reductions in accessory power. The developed framework is readily compatible with model predictive control and provides a template that, with site-specific recalibration, could be adapted to other liquid-cooled data centers with different configurations and cooling requirements.

eess.SY↗
arXiv2026

Co-Design Optimization for Data Center Cooling System via Digital Twin

Liquid-cooled exascale supercomputers dissipate heat through cooling plants organized as multiple parallel subloops, but how to allocate coolant distribution units (CDUs) across subloops and how to distribute flow among them has not been systematically addressed for facilities at this scale. This paper presents a three-layer optimization framework that jointly determines the integer partition of CDUs across subloops, the continuous flow fraction allocation, and the per-timestep co-design optimization of total flow rate and supply temperature subject to per-subloop thermal safety constraints. The Modelica simulation model is built based on the data of the Frontier exascale supercomputer at Oak Ridge National Laboratory. By developing a reduced-order surrogate model, all 611 feasible partitions of 25 CDUs are evaluated across the full year operational dataset of 49,353 timesteps. Three progressively richer operational strategies are compared, ranging from flow control optimization to full three-layer co-design optimization with dynamically adjusted flow fractions. The optimal design within the surrogate optimization problem is a two-subloop plant achieving 35.48% annual cooling energy savings, only 0.18% above the current three-subloop design at 35.30%. Most of the savings are delivered by supervisory co-optimization of total flow rate and supply temperature; the distinct role of flow fraction optimization is design robustness rather than additional raw savings. Flow fraction optimization compensates for any feasible CDU-to-subloop assignment, reducing the design sensitivity by 93% and providing a low-cost software-only pathway to near-optimal performance on the existing Frontier hardware. The framework is transferable to other liquid-cooled high-performance computing plants.

eess.SY↗
arXiv2026

Deltoris: Enabling Real-time VLA Inference in Embodied AI via Bit-level Sparsity and Speculative Inference

Vision-language-action (VLA) models have emerged as a key component in embodied AI. Among existing approaches, diffusion-based VLA models achieve superior motion quality and generalization. However, diffusion-based VLA models are compute-intensive and must run at high control frequency, e.g., 50-200 Hz. Thus, it imposes strict latency and energy constraints on edge devices. In this work, we present Deltoris, an algorithm-hardware co-design framework for efficient diffusion-based VLA inference. First, we exploit the temporal similarity of consecutive inputs and propose a \textit{temporal-aware bit-sparsity} algorithm that computes only the differences between consecutive inputs, eliminating redundant bit-level operations. To further address the extra off-chip traffic introduced by our algorithm, we propose a \textit{speculative inference} technique, which amortizes data loading across multiple control steps. Lastly, to support these techniques, we co-design a dedicated accelerator with customized 1D systolic bit-serial PE arrays that eliminate PE workload imbalance. Our evaluation shows that Deltoris achieves up to 34.2$\times$ speedup over mobile GPUs and 6.1$\times$ over prior accelerators, while maintaining comparable accuracy.

cs.AR↗
arXiv2026

Evaluating Counterfactual Sensitivity to Patient Information in Medication-Safety Reasoning

Applying a valid medication-safety rule when its patient-specific conditions are not met can produce an incorrect decision. Existing medical evaluations largely use isolated and fixed scenarios. A model may therefore answer correctly by recalling a drug-risk association without showing that it used patient information to decide whether the rule applies. To address this gap, we introduce MedPIC-Bench, a benchmark of source-verifiable recommendations and expert-validated questions for patient-specific medication-safety reasoning. It combines guideline-following questions with paired counterfactual questions in which a controlled change in patient information changes whether a rule applies. The benchmark contains 467 questions annotated along six clinical and reasoning dimensions. Across 28 medical-specific, general, and proprietary LLMs, every model performs worse on counterfactual questions, with mean accuracy falling from 63.6\% to 45.1\%. Models perform well when an explicit patient attribute directly signals a familiar contraindication, but struggle when patient information must narrow or withdraw a safety warning. Model rationales often acknowledge the changed patient information, yet the final answers retain the previous safety judgment. This vulnerability persists among medical-specific LLMs, whose average CF performance trails that of general LLMs. MedPIC-Bench therefore makes conditional rule application measurable and highlights the limitations of static medication-safety accuracy for assessing patient-specific reliability.

cs.AI↗
arXiv2026

Qwen-CUA: Native Computer Use for (almost) Everything

Native computer use offers a general interface for agents to operate almost any software available to people, but requires long-horizon state tracking, large-scale interactive experience, and learning from sparse yet verifiable outcomes. We introduce Qwen-CUA, a native computer-use agent with a 397B-A17B Qwen mixture-of-experts backbone. It observes only screenshots and acts through keyboard and mouse events, without DOM trees, accessibility metadata, or task-specific APIs. Its scaffold maintains up to 20 active screenshots and folds older visual history in fixed-size blocks to retain recent evidence while preserving reusable prompt prefixes. For training, we build a cloud rollout fleet with access to nearly 100,000 vCPUs and tens of thousands of concurrent environments, construct approximately 40,000 verifiable tasks, and collect personalized long-horizon workflows across everyday and professional software. We optimize complete trajectories with verifiable rewards and trajectory slicing, while iterative training runs refresh supervised data and recalibrate reinforcement-learning tasks. Across eight benchmarks, Qwen-CUA outperforms Qwen3.7 and remains competitive with leading proprietary systems, reaching 86.2 on OSWorld-Verified and 18.5/48.4 binary/partial completion on OSWorld 2.0. Scaling the same recipe to a model with over one trillion parameters yields Qwen-CUA-Max, improving these scores to 87.6 and 21.2/53.3. Qwen-CUA also reduces RedTeamCUA attack success from 36.6 to 16.4 relative to Qwen3.7. Efficiency analyses, a browser deployment, and Bash-augmented experiments further characterize practical behavior. These results establish native computer use as a broadly capable agent foundation and highlight scalable verifiable interaction and hybrid tool use as key directions.

cs.LG↗
arXiv2026

LiteResearcher: A Scalable Agentic RL Training Framework for Deep Research Agent

Reinforcement Learning (RL) has emerged as a powerful training paradigm for LLM-based agents. However, scaling agentic RL for deep research remains constrained by two coupled challenges: hand-crafted synthetic data fails to elicit genuine real-world search capabilities, and real-world search dependency during RL training introduces instability and prohibitive cost, which limits the scalability of Agentic RL. LiteResearcher is a training framework that makes Agentic RL scalable: by constructing a lite virtual world that mirrors real-world search dynamics, we enable a continuously improving training recipe that empowers a tiny search agent to outperform large-scale open-source and commercial models (e.g., Tongyi DeepResearch and Claude-4.5 Sonnet). Specifically, on common benchmarks such as GAIA and Xbench, our LiteResearcher-4B achieves open-source state-of-the-art results of 71.3% and 78.0% respectively, demonstrating that scalable RL training is a key enabler for Deep Research Agents.

cs.AI↗
arXiv2026

TENT: A Declarative Slice Spraying Engine for Performant and Resilient Data Movement in Disaggregated LLM Serving

Modern GPU clusters rely on complex, heterogeneous interconnects. As large language model (LLM) serving shifts toward agentic reasoning, KVCache becomes a first-class mobile asset, driving frequent migrations and massive elephant flows that dominate the execution critical path. Operating Mooncake Transfer Engine (TE) on thousands of GPUs exposed a fundamental flaw in existing frameworks: imperative, early-binding path selection. This rigidity results in state-blind striping that ignores congestion and grey failures, leading to bandwidth stranding. This also produces operational fragility where routine faults require manual intervention. We present TENT, a declarative orchestration engine that decouples transfer intent from physical execution. By abstracting interconnects into a unified resource pool, TENT shifts path resolution from initialization to slice-time late binding. Applications simply declare transfer intents, while TENT dynamically "sprays" fine-grained slices across rails based on real-time telemetry and predictive cost modeling. This orchestration eliminates head-of-line (HoL) blocking and enables transparent, sub-50 ms self-healing by rerouting slices around failures or degradations without application-level intervention. TENT serves as the production data plane for LLM inference and reinforcement learning (RL) pipelines at multiple industrial clusters. Our evaluation shows that TE+ outperforms state-of-the-art baselines, including Mooncake TE, NIXL, and UCCL. In LLM inference with SGLang HiCache, TE+ achieves up to 1.36x higher throughput and 26% lower P90 time-to-first-token (TTFT) than Mooncake TE. In RL pipelines, TENT accelerates parameter updates in Moonshot Checkpoint Engine by 20-26%.

cs.DC↗
arXiv2026

Trajectory-Aware Information Matching for Multi-Step Gradient Inversion in Federated Learning

Federated learning enables distributed information sharing and collaborative model training without exposing raw client data. However, shared gradients or model updates may still contain sensitive information, making federated learning vulnerable to gradient inversion attacks. Most existing gradient inversion attacks rely on simplified update observations, such as single-step gradients or endpoint-based matching. In practical FL, however, FedAvg produces an accumulated trajectory-dependent update after multiple local steps, rather than a gradient computed at a single model state.To address this issue, we propose NL-SME, a trajectory-aware information matching method for multi-step gradient inversion. NL-SME constructs a learnable nonlinear surrogate trajectory to approximate hidden local states and integrates trajectory-level information with calibrated gradient matching. For perturbed updates, NL-SME can further use an observed-update reliability-aware strategy to reduce the influence of unreliable components. Extensive experiments under diverse multi-step FedAvg settings show that NL-SME outperforms state-of-the-art gradient inversion baselines in reconstruction quality and update-matching accuracy. Additional evaluations on natural and medical images, as well as under fused-update observations and representative defense strategies, further suggest that observable multi-step updates may still retain reconstruction signals. These results reveal potential privacy leakage risks in federated information sharing. Code is available at https://anonymous.4open.science/r/NL-SME-main/README.md.

cs.LG↗
arXiv2026

Physics-Grounded Monocular Vehicle Distance Estimation Using Standardized License Plate Typography

Accurate inter-vehicle distance estimation is a cornerstone of Advanced Driver Assistance Systems (ADAS) and autonomous driving. While LiDAR and radar provide high precision, their high cost prohibits widespread adoption in mass-market vehicles. Monocular camera-based estimation offers a low-cost alternative but suffers from fundamental scale ambiguity. Recent deep learning methods for monocular depth achieve impressive results yet require expensive supervised training, suffer from domain shift, and produce predictions that are difficult to certify for safety-critical deployment. This paper presents a framework that exploits the standardized typography of United States license plates as passive fiducial markers for metric ranging, resolving scale ambiguity through explicit geometric priors without any training data or active illumination. First, a four-method parallel plate detector achieves robust plate reading across the full automotive lighting range. Second, a three-stage state identification engine fusing optical character recognition text matching, multi-design color scoring, and a lightweight neural network classifier provides robust identification across all ambient conditions. Third, hybrid depth fusion with inverse-variance weighting and online scale alignment, combined with a one-dimensional constant-velocity Kalman filter, delivers smoothed distance, relative velocity, and time-to-collision for collision warning. Baseline validation on a controlled static dataset reproduces a 2.3% coefficient of variation in character height measurements and a 36% reduction in distance-estimate variance compared with plate-width methods from prior work.

cs.CV↗
arXiv2026

Controlled chemical vapor deposition for synthesis of emerging Mo(W)Te2 systems

The Group-VI transition metal ditellurides offer a rich platform for correlated and topological phenomena, yet their structural polymorphism and instability complicate the creation of single crystals and heterointerfaces. Here, we introduce a confined-space chemical vapor deposition (CVD) strategy that lowers the growth temperature window and, when combined with tailored precursor configurations and stepwise thermal ramps, enables the deterministic synthesis of high-quality single crystals, alloys, and lateral/vertical heterostructures. High-resolution aberration-corrected STEM provides atomic characterization of lattice-matched Mo(W)Te2 lateral heterostructure, revealing nearly atomically sharp, compositionally well-defined seamless boundaries. This approach avoids the thickness nonuniformity and structural limitations commonly associated with exfoliated samples, enabling reproducible fabrication of clean heterointerfaces and establishing a nearly ideal in-situ experimental system. Furthermore, scanning tunneling microscopy and spectroscopy (STM and STS) enable direct imaging of the seamless boundaries in Mo(W)Te2 lateral heterostructures, while uncovering their distinct real-space distributions of the local density of states. Our results establish a scalable pathway for engineering crystalline Te-based structures with controlled geometry and stacking, providing an essential step toward quantum and topological device platforms based on the transition metal ditellurides family.

cond-mat.mtrl-sci↗
arXiv2026

Structural Assessment for Understanding and Guiding Dataset Distillation in Discrete Token Space

Dataset distillation (DD) has proven to reduce training cost while preserving accuracy. While promising, the factors that make one distilled dataset more effective than another remain poorly understood. In this work, we investigate this question through the lens of discrete visual tokenizers. Whereas many prior DD efforts emphasize matching global data distributions, we suggest that the effectiveness depends on which semantic concepts are captured and how they are composed. Discrete visual tokenizers provide a finite vocabulary that enables direct statistical analysis of such compositional structure. Through quantitative analysis of token-level statistics, we introduce the structural score to measure the adequacy of token compositions. We observe that distilled datasets with balanced token composition yield higher validation performance. On the other hand, divergence from the original data does not necessarily harm performance. We further show that samples with high structural scores in the discrete token space can effectively guide diffusion-based DD. Our findings highlight the importance of token composition in dataset effectiveness, offering a principled complement to distributional similarity considerations in DD.

cs.CV↗
arXiv2026

SAC: Disaggregated KV Cache System for Sparse Attention LLMs with CXL

The scaling of LLMs toward long-context inference has shifted the primary serving system bottleneck from computation to memory capacity. Traditional solutions for dense attention models rely on RDMA-based disaggregated memory pools, which perform coarse-grained fetching of the entire prefix KV cache from remote storage to local memory before decoding. However, this approach is fundamentally inefficient for emerging sparse attention models. While only a small fraction of KV entries are active during decoding, these systems still fetch the full KV cache locally, leading to severe transmission bottlenecks and local memory wastage. To address this, we propose SAC, the first efficient disaggregated KV cache system optimized for sparse attention models. By leveraging the low-latency, cache-line granularity load/store semantics of Compute Express Link (CXL), SAC fetches only the required top-k KV entries on demand during inference. Evaluations on DeepSeek-V3.2 using SGLang show that SAC achieves 2.1x higher throughput, 9.7x lower TTFT, and 1.8x lower TBT compared to RDMA-based baselines, establishing CXL-based disaggregation as the superior infrastructure for emerging sparse attention models.

cs.DC↗
arXiv2026

A Dual-Branch Collaborative Framework for Joint Optimization of Underwater Image Enhancement and Object Detection

Due to wavelength dependent light absorption and scattering, underwater images usually suffer from color distortion and blurred details, which limits underwater object detection performance. Existing underwater image enhancement methods mainly focus on visual quality improvement, while it is still difficult to balance enhancement quality, processing efficiency, and downstream detection performance. Therefore, this paper proposes an efficient dual-branch underwater image enhancement framework for object detection. The detail enhancement branch improves brightness and local contrast to recover texture details in dark regions. The color restoration branch uses adaptive compensation to reduce color distortion and improve color gradation. By combining the complementary outputs of the two branches, the proposed framework provides clearer and more informative images for object detection. On the UIEB and EUVP datasets, the proposed method achieves UIQM scores of 2.249 and 2.576. When applied to the YOLOv8 detection task on the URPC dataset, the proposed method improves mAP50 by 2.1\% compared with the baseline. Extensive experiments show that our method improves object detection in complex underwater scenes, while balancing enhancement quality and processing efficiency.

cs.CV↗
arXiv2026

Pronounced in-plane anomalous Hall effect with vanishing out-of-plane response in Cr1.2Te2

We report an unconventional anomalous Hall regime in the van der Waals ferromagnet Cr1.2Te2, in which the anomalous Hall effect (AHE) is present for in-plane magnetization but absent for out-of-plane magnetization. In this purely in-plane regime, the anomalous Hall signal exhibits a threefold angular dependence during both in-plane and out-of-plane rotations of the magnetization, which cannot be accounted for by the conventional dipolar contribution but instead requires an octupolar contribution. Although the octupolar term qualitatively captures the observed behavior, the experimentally extracted octupole differs quantitatively from first-principles calculations based solely on the intrinsic Berry-curvature mechanism, indicating an essential role for extrinsic scattering processes.

cond-mat.mtrl-sci↗
arXiv2026

PrivCode++: Latent-Conditioned Differentially Private Code Generation for Comprehensive Guarantees

Large language models fine-tuned on instruction-code pairs may memorize and subsequently leak sensitive training data. Existing differentially private (DP) code generation methods primarily protect code snippets while assuming prompts are public, which fails in realistic scenarios where prompts may also contain sensitive information. When prompts cannot be explicitly learned or used during generation, code synthesis suffers from severe utility degradation as well as reduced diversity and fidelity. To address these challenges, we propose PrivCode-Plus, the first work to explore DP code generation where both prompts and code snippets are considered sensitive in LLM fine-tuning. PrivCode-Plus introduces a two-stage DP framework with a Privacy-Free Latent Conditioning module, enabling effective DP fine-tuning and data synthesis without direct access to sensitive prompts or code. Extensive experiments show that PrivCode-Plus achieves substantially higher utility than baselines, remains competitive with the method with relaxing privacy assumptions, and provides stronger privacy guarantees.

cs.CR↗
arXiv2026

AutoDFT: A Closed-Loop Multi-Agent Framework for Autonomous DFT Calculations

Density functional theory (DFT) serves as the basis for computational discovery in materials science and chemistry, yet each calculation demands extensive human effort: adjusting algorithms when convergence stalls, revising plans when unexpected physics emerges, and inserting steps as intermediate results reshape the problem. Existing LLM-based agents automate only the initial planning stage, producing a full execution plan upfront and leaving all subsequent adaptation to hand-crafted rules. As a result, these workflows remain fragile, do not generalize well beyond pre-planned scenarios, and often require expert intervention when failures or unexpected intermediate results require changes to the calculation path. Here, we introduce AutoDFT, a closed-loop multi-agent framework that embeds LLM reasoning into every stage of the DFT lifecycle, where a strategic planner produces a skeletal plan of step objectives; a step planner generates numerical parameters just in time from preceding results; and a monitor-recover-reflect cycle diagnoses failures, repairs them, and revises the plan when the evidence justifies it. We demonstrate both breadth and depth: breadth on VASPBench, a purpose-built benchmark spanning 34 tasks and 9 DFT calculation types, where AutoDFT achieves 94.1% task-level success with GPT-5.2; and depth on established materials databases, where AutoDFT produces quantitatively reliable property predictions across electronic, magnetic, and energetic properties. By closing the loop between planning and execution, AutoDFT enables experimentalists without deep computational expertise to obtain reliable first-principles results.

cond-mat.mtrl-sci↗
arXiv2026

Cooling Channel Design Optimization for High Power Multi-Chip Packages

Thermal management is a major challenge in next-generation high-performance computing systems, particularly for heterogeneous multi-chip packages such as the NVIDIA GB200 Grace Blackwell Superchip. In this work, a physics-based computational framework is developed to optimize embedded cooling channel layouts for high-power multi-chip modules. The model couples steady-state heat conduction with a porous media-based representation of coolant transport, coupled with a row-wise coolant energy balance, to estimate chip temperature fields within microchannel networks. Unlike conventional designs, an interdigitated cooling architecture is parameterized using geometric variables, including channel count, width, and expansion over chip regions, enabling systematic design exploration. To enable efficient optimization, a surrogate-based approach is employed to approximate the relationship between geometric parameters and temperature metrics. The resulting model is optimized using a mixed-integer quadratic programming algorithm to minimize a weighted objective based on peak and average chip temperatures. To improve physical relevance, channel placement is further constrained to increase cooling coverage near GPU regions, where thermal loads are highest. The framework is applied to a representative multi-chip configuration based on NVIDIA GB200 architecture, consisting of two graphics processing units and one central processing unit. The results demonstrate that the optimal design reduces the peak chip temperature by 140.45°C and the average chip temperature by 35.87°C compared to the baseline configuration.

eess.SY↗
arXiv2026

ARBOR: Online Process Rewards via a Reusable Rubric Buffer for Search Agents

LLM-based search agents are trained predominantly with outcome-only reward, leaving the search process itself unsupervised. This signal degenerates on outcome-homogeneous groups where all sampled trajectories share the same correctness, yielding zero within-group advantage and no gradient. Existing process supervision either trains a costly verifier or generates per-query rubrics that are inconsistent across queries and discarded after one use. We propose ARBOR (Adaptive Rubric Buffer for Online Reward), a reusable process-reward framework that maintains a rubric memory shared across queries. Query-local drafts induced from contrastive trajectories are admitted, consolidated into cross-query common rubrics, and retired as the policy evolves. A small active subset of common rubrics scores trajectories via sparse pairwise judging, and the resulting scores are added to the base reward, providing process-level gradient even when outcome reward is uniform. ARBOR consistently outperforms GRPO and DAPO baselines on four multi-hop QA benchmarks, raising average LLM-judge accuracy by up to 4.2 points and converting up to 42% of otherwise-zero-gradient training groups into informative ones.

cs.CL↗