arXiv subjects

Jinghao Wang

Publications and source records attributed to Jinghao Wang.

At least 19 recordsLinked to original sources

arXiv2026

Latency-Aware Orchestration for Multi-Agent LLM Workflows on Heterogeneous GPUs

Concurrent multi-agent workflows expose future dependencies and serving-state requirements while running on heterogeneous GPU pools with time-varying load, model residency, and resource availability. The logical workflow defines the required computation, whereas its physical scheduling units, model-lifecycle actions, resource ordering, and placement must be selected according to the observed pool state. We present a prediction-guided runtime that uses workflow forecasts to construct and optimize a physical execution graph. Predictor estimates device-specific activation latency, peak memory, and model-loading cost, then propagates these predictions through workflow dependencies to forecast activation readiness and future model demand. Constructor builds semantics-preserving fusion and model-lifecycle alternatives, while Scheduler jointly optimizes their selection, placement, and execution order based on the live pool state. Across a workload spanning three workflow scenarios on a heterogeneous GPU pool, our system reduces end-to-end makespan and overall p95 completion latency under burst arrivals by up to 36.8% and 25.9%, respectively, over state-of-the-art workflow schedulers. It also saves up to 24.63 GPU-s per completed session.

cs.DC↗

arXiv2026

HIERA: Workload-Aware Planning Across Implementation Spaces for GPU Kernel Optimization

High-performance GPU kernels underpin modern deep learning and scientific computing. As workloads become increasingly diverse and GPU hardware evolves rapidly, developing efficient methods for automated GPU kernel generation and optimization has become increasingly important. Existing LLM-based methods typically optimize within a fixed implementation space, limiting either optimization flexibility or search efficiency. We propose \textsc{HIERA}, a hierarchical search-space planning framework for GPU kernel optimization. \textsc{HIERA} constructs contract-augmented task specifications, selects an appropriate implementation space across PyTorch operators, CUDA libraries, and custom CUDA kernels, and uses profiling feedback and expert knowledge to guide structured iterative refinement. Experiments on KernelBench across multiple various workload levels and base LLMs show that \textsc{HIERA} delivers stronger overall implementation validity, sample efficiency, and optimization performance than existing training-free methods, while remaining competitive with the training-based CUDA-L1 without additional model training. A case study on a specialized stencil operator from scientific computing further achieves a $1.53\times$ speedup over cuDNN, demonstrating the potentiality of the general framework beyond standard machine-learning workloads.

cs.DC↗

arXiv2026

ElastiCo: Elastic Configuration and Interference-Aware Orchestration for GPU Clusters

Modern GPU clusters must simultaneously serve deep learning training and offline large language model inference workloads, yet existing schedulers treat these as isolated resource consumers with rigid, static allocations. This leaves substantial GPU capacity underutilized: training jobs reserve entire devices despite periodic idle phases, while offline inference tasks over-provision GPUs despite bursty demand patterns. We present ElastiCo, an elastic co-location framework that enables training and inference workloads to safely share GPUs through three integrated mechanisms. First, Resource Shape Transformation exposes each job as a family of feasible resource-performance configurations. Second, Elastic Shadow Pricing decomposes the resulting multi-resource allocation problem into per-job configuration selection subproblems via dynamic per-resource shadow prices. Third, Interference-Aware Co-location uses a predictor trained on hardware-counter and task-level features to estimate pairwise performance degradation under GPU sharing. Implemented as native Kubernetes middleware requiring no user-code modifications, ElastiCo is evaluated on a 64-GPU testbed and through large-scale trace-driven simulations (up to 512 GPUs), reducing the average JCT by up to 2.94x, increasing the cluster throughput by 2.02x, and increasing the GPU utilization from approximately 25% to 46%.

cs.DC↗

arXiv2026

SpecBox: Speculative Sandbox Scheduling for Efficient LLM Agent Serving

As LLM agents increasingly rely on the Model Context Protocol (MCP) to invoke isolated external sandboxes, disaggregated sandbox deployment introduces a fundamental tension between resource utilization and interactive tail latency. Persistent long-lived sandbox reservations incur excessive memory overhead at scale, while lazy on-demand instantiation generates severe cold-start penalties that degrade response performance under multi-tenant, multi-turn agent workloads. To resolve this dilemma, we present SpecBox, a runtime built around speculative sandbox preallocation tailored for dynamic LLM agent execution pipelines. At its core, SpecBox implements keyword matching and streaming semantic embedding to enable intent-driven sandbox prewarming, which identifies pending tool execution demands mid-LLM token generation and fully overlaps sandbox bootstrapping with model inference. To extend prewarming windows across sequential agent steps, the framework leverages context-aware stochastic prefetching atop a sandbox dependency graph to probabilistically forecast future sandbox switches ahead of execution. We complement these speculative mechanisms with two orthogonal optimizations: a semantic result cache that prunes redundant repeated sandbox invocations, and a dedicated out-of-band shared-memory transport plane that bypasses conventional network serialization to deliver zero-copy artifact transfers. Evaluated on high-concurrency multi-turn agent traces, our prototype demonstrates that SpecBox cuts P99 end-to-end latency by up to $2.9\times$ relative to the on-demand sandbox baseline, while slashing peak memory consumption by $45.9\%$ compared to permanently reserved sandbox deployments.

cs.DC↗

arXiv2026

Learning Cross-View Semantic Priors for Single-Reference Unseen Object Pose Estimation

Single-reference unseen object 6D pose estimation reduces object onboarding by estimating poses of arbitrary novel objects from only one reference view. Recent correspondence-based pipelines have achieved robust performance with vision foundation model (VFM) features. However, they typically treat these features as intra-view descriptors, leaving dense visual-semantic cues, including appearance, structure, and context, insufficiently exchanged across views before geometric decoding. Consequently, the decoded point features may lack joint semantic and geometric discriminability, making correspondence estimation still difficult in challenging cases. Instead of processing features independently, we build the correspondence pipeline around an early cross-view semantic prior. Specifically, cross-view semantic interaction (CVSI) enables dense query and reference VFM tokens to exchange semantic context and form a cross-view prior. Nevertheless, direct CVSI may disturb the VFM token structure, while the resulting semantic prior still needs 3D representation consistency for rigid correspondence. To make this CVSI prior reliable for 3D correspondence learning, we introduce two complementary training-time constraints: the intra-view structure preservation (IVSP) loss preserves the original intra-view token affinity structure during interaction, while the reference-anchored geometric consistency (RAGC) loss enforces spatial representation consistency of decoded point features. The final pose is recovered from learned correspondences through weighted SVD. We further construct a challenging view-pair protocol from the BOP Challenge datasets YCB-V and TUD-L to evaluate robustness in difficult matching scenarios. Extensive experiments on six benchmarks under different view-pair settings show that our method achieves state-of-the-art performance while maintaining comparable inference speed.

cs.CV↗

arXiv2026

Band-minimum degeneracy is not enough: density-of-states control of low-density ferromagnetism

Recent ultracold-atom experiments have observed ferromagnetic correlations in a geometrically tunable Fermi--Hubbard lattice at intermediate densities. Motivated by subsequent theoretical work that connected the low-density limit of the same model to the Mueller--Hartmann mechanism for itinerant ferromagnetism, we investigate the stability of ferromagnetism across a broad density range using the T-matrix approximation, a controlled approach in the dilute limit. We reproduce the ferromagnetic regime observed experimentally at intermediate densities and systematically compare finite-size and thermodynamic-limit calculations in the dilute regime. We find that the low-density ferromagnetic phase reported for the one-diagonal hopping lattice is strongly suppressed with increasing system size and disappears in the thermodynamic limit, indicating that it originates from finite-size effects. By contrast, low-density ferromagnetism remains robust in the square-lattice Hubbard model with hopping along both diagonal directions. We show that this qualitative difference cannot be explained by the degeneracy of the band minima alone, which occurs in both models. Instead, it is controlled by the singular behavior of the density of states at the band bottom: a weak divergence is insufficient to stabilize ferromagnetism, whereas a much stronger quasi-one-dimensional singularity supports a fully polarized ground state even in the dilute limit. Our results demonstrate that band-minimum degeneracy alone is not sufficient for low-density ferromagnetism and that the nature of the band-bottom density-of-states singularity ultimately controls its stability.

cond-mat.quant-gas↗

arXiv2026

Maestro: Workload-Aware Cross-Cluster Scheduling for LLM-Based Multi-Agent Systems

Large Language Model based Multi-Agent Systems (LLM-MAS) have emerged as a powerful paradigm for tackling complex tasks by breaking them into collaborative workflows of specialized LLM-powered agents. However, deploying such multi-agent workloads at scale poses significant system challenges. Each user query spawns an iterative pipeline of LLM calls, greatly amplifying resource consumption compared to single-turn queries. In resource-constrained cloud settings, these workflows face non-deterministic and input-dependent costs at decode stage, heavy-tailed multi-model requirements with memory fragmentation and over-provisioning, and cross-cluster scheduling trade-offs. We present Maestro, a workload-aware scheduling system designed for LLM-MAS serving under strict GPU budgets. Maestro explicitly leverages agent semantics and roles: it predicts the output length and memory usage of each stage and uses this prediction to drive a hierarchical scheduler. At the node level, Maestro enables dynamic multi-model co-location via hierarchical weight caching and elastic memory provisioning. At the cluster level, it performs latency-aware routing to avoid cold-start delays and memory overloads. At the global level, it enforces workflow-aware prioritization to minimize head-of-line blocking for interactive tasks. Across prototype experiments and trace-driven simulations, Maestro reduces KV-reservation HBM by 67.2% and improves high-contention SLO attainment over EDF by 23.6 percentage points.

cs.DC↗

arXiv2026

TravExplorer: Cross-Floor Embodied Exploration via Traversability-Aware 3-D Planning

Zero-shot Object Navigation (ZSON) has shown promise for open-vocabulary target search in unseen environments, yet most existing systems remain tied to planar representations and single-floor assumptions. These assumptions become inadequate in real buildings, where navigation involves floors, stairs, landings, and vertically overlapping spaces. This article presents TravExplorer, a cross-floor embodied exploration framework that couples zero-shot semantic guidance with traversability-aware 3-D planning. TravExplorer maintains a unified volumetric map that distinguishes occupied structures from robot-reachable support surfaces and extracts traversable frontiers from connected support surfaces, including floors, stairs, and landings. A FOV-aware active perception strategy further resolves incomplete observations during cross-floor traversal. To reduce semantic-reasoning latency, a lightweight guidance module aligns a probabilistic instance map from online open-vocabulary segmentation with a spatial value map from fast image-to-text matching. Based on these geometric and semantic memories, a hierarchical planner performs target-aware frontier touring over object hypotheses, traversable frontiers, and stair landmarks, and generates executable cross-floor motions through foothold-guided 3-D search and vertically constrained local trajectory optimization. Experiments over 4,195 simulated episodes on HM3D and MP3D demonstrate consistent advantages over representative ObjectNav baselines. Fifty real-world trials on a Unitree Go2 further validate open-vocabulary target search across single-floor and cross-floor indoor environments without prior maps or human intervention. The code will be released at https://github.com/wuyi2121/TravExplorer.

cs.RO↗

arXiv2026

Does Engram Do Memory Retrieval in Autoregressive Image Generation?

The Engram module -- a hash-keyed, O(1) associative memory injected into Transformer layers -- was recently shown to improve large language model pretraining, with the appealing interpretation that it provides a content-addressed shortcut to recurring local token patterns. We ask whether this interpretation transfers to autoregressive (AR) image generation, or whether the observed gains, if any, come from a different mechanism. We adapt the Engram module to vision with 2D spatial $n$-gram hashing, gated fusion, and KV-cache-compatible incremental inference, and inject it into a class-conditional AR generator trained on ImageNet 256x256. Across a sweep of backbone-to-memory budget ratios $\rho{\in}[0.17, 0.90]$, every Engram-augmented variant trails the pure AR baseline in FID, indicating that the module saves backbone FLOPs but does not, by itself, improve sample quality. We then probe how the module is used. A gate-clamp sweep shows that disabling the Engram pathway entirely is catastrophic, yet a tiny constant gate (g=0.10) matches or beats the learned gate -- inconsistent with a heavily content-addressed recall mechanism. A donor-probe experiment shows that swapping the hash inputs for matched, adversarial, or random same-class exemplars produces statistically indistinguishable next-token distributions, while collapsing or randomising the table degrades them by two to three orders of magnitude. Finally, training a model from scratch with the entire memory table frozen to $\mathcal{N}(0, 1)$ noise costs only $\Delta\text{FID}{=}0.10$ and actually raises Inception Score. Together, these findings indicate that the Engram in AR image generation behaves not as a content-addressed retriever but as a gated architectural side-pathway: a hash-keyed residual stream whose benefit is dominated by the pathway itself, with the learned table contributing only a small distributional refinement.

cs.CV↗

arXiv2025

A Practical Framework for Evaluating Medical AI Security: Reproducible Assessment of Jailbreaking and Privacy Vulnerabilities Across Clinical Specialties

Medical Large Language Models (LLMs) are increasingly deployed for clinical decision support across diverse specialties, yet systematic evaluation of their robustness to adversarial misuse and privacy leakage remains inaccessible to most researchers. Existing security benchmarks require GPU clusters, commercial API access, or protected health data -- barriers that limit community participation in this critical research area. We propose a practical, fully reproducible framework for evaluating medical AI security under realistic resource constraints. Our framework design covers multiple medical specialties stratified by clinical risk -- from high-risk domains such as emergency medicine and psychiatry to general practice -- addressing jailbreaking attacks (role-playing, authority impersonation, multi-turn manipulation) and privacy extraction attacks. All evaluation utilizes synthetic patient records requiring no IRB approval. The framework is designed to run entirely on consumer CPU hardware using freely available models, eliminating cost barriers. We present the framework specification including threat models, data generation methodology, evaluation protocols, and scoring rubrics. This proposal establishes a foundation for comparative security assessment of medical-specialist models and defense mechanisms, advancing the broader goal of ensuring safe and trustworthy medical AI systems.

cs.CR↗

arXiv2025

DKPMV: Dense Keypoints Fusion from Multi-View RGB Frames for 6D Pose Estimation of Textureless Objects

6D pose estimation of textureless objects is valuable for industrial robotic applications, yet remains challenging due to the frequent loss of depth information. Current multi-view methods either rely on depth data or insufficiently exploit multi-view geometric cues, limiting their performance. In this paper, we propose DKPMV, a pipeline that achieves dense keypoint-level fusion using only multi-view RGB images as input. We design a three-stage progressive pose optimization strategy that leverages dense multi-view keypoint geometry information. To enable effective dense keypoint fusion, we enhance the keypoint network with attentional aggregation and symmetry-aware training, improving prediction accuracy and resolving ambiguities on symmetric objects. Extensive experiments on the ROBI dataset demonstrate that DKPMV outperforms state-of-the-art multi-view RGB approaches and even surpasses the RGB-D methods in the majority of cases. The code will be available soon.

cs.CV↗

arXiv2025

REAR: Rethinking Visual Autoregressive Models via Generator-Tokenizer Consistency Regularization

Visual autoregressive (AR) generation offers a promising path toward unifying vision and language models, yet its performance remains suboptimal against diffusion models. Prior work often attributes this gap to tokenizer limitations and rasterization ordering. In this work, we identify a core bottleneck from the perspective of generator-tokenizer inconsistency, i.e., the AR-generated tokens may not be well-decoded by the tokenizer. To address this, we propose reAR, a simple training strategy introducing a token-wise regularization objective: when predicting the next token, the causal transformer is also trained to recover the visual embedding of the current token and predict the embedding of the target token under a noisy context. It requires no changes to the tokenizer, generation order, inference pipeline, or external models. Despite its simplicity, reAR substantially improves performance. On ImageNet, it reduces gFID from 3.02 to 1.86 and improves IS to 316.9 using a standard rasterization-based tokenizer. When applied to advanced tokenizers, it achieves a gFID of 1.42 with only 177M parameters, matching the performance with larger state-of-the-art diffusion models (675M).

cs.CV↗

arXiv2025

FreeMorph: Tuning-Free Generalized Image Morphing with Diffusion Model

We present FreeMorph, the first tuning-free method for image morphing that accommodates inputs with different semantics or layouts. Unlike existing methods that rely on finetuning pre-trained diffusion models and are limited by time constraints and semantic/layout discrepancies, FreeMorph delivers high-fidelity image morphing without requiring per-instance training. Despite their efficiency and potential, tuning-free methods face challenges in maintaining high-quality results due to the non-linear nature of the multi-step denoising process and biases inherited from the pre-trained diffusion model. In this paper, we introduce FreeMorph to address these challenges by integrating two key innovations. 1) We first propose a guidance-aware spherical interpolation design that incorporates explicit guidance from the input images by modifying the self-attention modules, thereby addressing identity loss and ensuring directional transitions throughout the generated sequence. 2) We further introduce a step-oriented variation trend that blends self-attention modules derived from each input image to achieve controlled and consistent transitions that respect both inputs. Our extensive evaluations demonstrate that FreeMorph outperforms existing methods, being 10x ~ 50x faster and establishing a new state-of-the-art for image morphing.

cs.CV↗

arXiv2025

Deterministic Object Pose Confidence Region Estimation

6D pose confidence region estimation has emerged as a critical direction, aiming to perform uncertainty quantification for assessing the reliability of estimated poses. However, current sampling-based approach suffers from critical limitations that severely impede their practical deployment: 1) the sampling speed significantly decreases as the number of samples increases. 2) the derived confidence regions are often excessively large. To address these challenges, we propose a deterministic and efficient method for estimating pose confidence regions. Our approach uses inductive conformal prediction to calibrate the deterministically regressed Gaussian keypoint distributions into 2D keypoint confidence regions. We then leverage the implicit function theorem to propagate these keypoint confidence regions directly into 6D pose confidence regions. This method avoids the inefficiency and inflated region sizes associated with sampling and ensembling. It provides compact confidence regions that cover the ground-truth poses with a user-defined confidence level. Experimental results on the LineMOD Occlusion and SPEED datasets show that our method achieves higher pose estimation accuracy with reduced computational time. For the same coverage rate, our method yields significantly smaller confidence region volumes, reducing them by up to 99.9\% for rotations and 99.8\% for translations. The code will be available soon.

cs.CV↗

arXiv2025

STRCMP: Integrating Graph Structural Priors with Language Models for Combinatorial Optimization

Combinatorial optimization (CO) problems, central to operation research and theoretical computer science, present significant computational challenges due to their NP-hard nature. While large language models (LLMs) have emerged as promising tools for CO--either by directly generating solutions or synthesizing solver-specific codes--existing approaches often neglect critical structural priors inherent to CO problems, leading to suboptimality and iterative inefficiency. Inspired by human experts' success in leveraging CO structures for algorithm design, we propose STRCMP, a novel structure-aware LLM-based algorithm discovery framework that systematically integrates structure priors to enhance solution quality and solving efficiency. Our framework combines a graph neural network (GNN) for extracting structural embeddings from CO instances with an LLM conditioned on these embeddings to identify high-performing algorithms in the form of solver-specific codes. This composite architecture ensures syntactic correctness, preserves problem topology, and aligns with natural language objectives, while an evolutionary refinement process iteratively optimizes generated algorithm. Extensive evaluations across Mixed Integer Linear Programming and Boolean Satisfiability problems, using nine benchmark datasets, demonstrate that our proposed STRCMP outperforms five strong neural and LLM-based methods by a large margin, in terms of both solution optimality and computational efficiency. The code and learned model will be publicly available upon the acceptance of the paper.

cs.LG↗

arXiv2024

Efficient Influence Minimization via Node Blocking

Given a graph G, a budget k and a misinformation seed set S, Influence Minimization (IMIN) via node blocking aims to find a set of k nodes to be blocked such that the expected spread of S is minimized. This problem finds important applications in suppressing the spread of misinformation and has been extensively studied in the literature. However, existing solutions for IMIN still incur significant computation overhead, especially when k becomes large. In addition, there is still no approximation solution with non-trivial theoretical guarantee for IMIN via node blocking prior to our work. In this paper, we conduct the first attempt to propose algorithms that yield data-dependent approximation guarantees. Based on the Sandwich framework, we first develop submodular and monotonic lower and upper bounds for our non-submodular objective function and prove the computation of proposed bounds is \#P-hard. In addition, two advanced sampling methods are proposed to estimate the value of bounding functions. Moreover, we develop two novel martingale-based concentration bounds to reduce the sample complexity and design two non-trivial algorithms that provide (1-1/e-ε)-approximate solutions to our bounding functions. Comprehensive experiments on 9 real-world datasets are conducted to validate the efficiency and effectiveness of the proposed techniques. Compared with the state-of-the-art methods, our solutions can achieve up to two orders of magnitude speedup and provide theoretical guarantees for the quality of returned results.

cs.DB↗

arXiv2024

AID: Attention Interpolation of Text-to-Image Diffusion

Conditional diffusion models can create unseen images in various settings, aiding image interpolation. Interpolation in latent spaces is well-studied, but interpolation with specific conditions like text or poses is less understood. Simple approaches, such as linear interpolation in the space of conditions, often result in images that lack consistency, smoothness, and fidelity. To that end, we introduce a novel training-free technique named Attention Interpolation via Diffusion (AID). Our key contributions include 1) proposing an inner/outer interpolated attention layer; 2) fusing the interpolated attention with self-attention to boost fidelity; and 3) applying beta distribution to selection to increase smoothness. We also present a variant, Prompt-guided Attention Interpolation via Diffusion (PAID), that considers interpolation as a condition-dependent generative process. This method enables the creation of new images with greater consistency, smoothness, and efficiency, and offers control over the exact path of interpolation. Our approach demonstrates effectiveness for conceptual and spatial interpolation. Code and demo are available at https://github.com/QY-H00/attention-interpolation-diffusion.

cs.CV↗

arXiv2023

Pair then Relation: Pair-Net for Panoptic Scene Graph Generation

Panoptic Scene Graph (PSG) is a challenging task in Scene Graph Generation (SGG) that aims to create a more comprehensive scene graph representation using panoptic segmentation instead of boxes. Compared to SGG, PSG has several challenging problems: pixel-level segment outputs and full relationship exploration (It also considers thing and stuff relation). Thus, current PSG methods have limited performance, which hinders downstream tasks or applications. The goal of this work aims to design a novel and strong baseline for PSG. To achieve that, we first conduct an in-depth analysis to identify the bottleneck of the current PSG models, finding that inter-object pair-wise recall is a crucial factor that was ignored by previous PSG methods. Based on this and the recent query-based frameworks, we present a novel framework: Pair then Relation (Pair-Net), which uses a Pair Proposal Network (PPN) to learn and filter sparse pair-wise relationships between subjects and objects. Moreover, we also observed the sparse nature of object pairs for both Motivated by this, we design a lightweight Matrix Learner within the PPN, which directly learns pair-wised relationships for pair proposal generation. Through extensive ablation and analysis, our approach significantly improves upon leveraging the segmenter solid baseline. Notably, our method achieves over 10\% absolute gains compared to our baseline, PSGFormer. The code of this paper is publicly available at https://github.com/king159/Pair-Net.

cs.CV↗