arXiv subjects

Hui Zhang

Publications and source records attributed to Hui Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Positive sectional curvature and non-isometric circle actions on a family of eleven-spheres

We study the construction of positively curved Riemannian metrics by non-isometric circle actions. For a family of homotopy eleven-spheres whose Eells--Kuiper invariants form the even subgroup of $\mathbb{Z}/992$, we construct, on each member, a smooth background metric $q$ and three effective circle actions with generators $W_1,W_2,W_3$ such that the metric determined by $g^{-1}=q^{-1}+\sum_{a=1}^3W_a\otimes W_a$ has positive sectional curvature. Each action is non-isometric for every partial metric, including its incoming metric and the final metric. We first describe the sphere by gauge transformations of the quaternionic Hopf bundle. We then construct compatible metrics on two disks and smooth their inverse metrics while preserving the action formula. A local conjugation makes the circle actions non-isometric. For each fixed member, we obtain an explicit positive lower bound $2^{-54}(1+M_{0,k}+M_{1,k})^{-28}$, where $M_{0,k}$ and $M_{1,k}$ are norms of the curvature and its first covariant derivative for its fixed connection. The bound may depend on the member of the family.

math.DG↗

arXiv2026

LoongSage: A Production-Grade, High-Efficiency Agentic RL Framework for Frontier LLMs with Validated Recipes

Reinforcement learning (RL) has emerged as a critical phase in the post-training pipeline of large language models (LLMs). As the frontier of LLM capabilities shifts from static reasoning to agentic tasks--characterized by external tool invocation, multi-turn interactions, and long-horizon decision-making--existing mainstream RL frameworks have grown increasingly unwieldy, raising the overhead of adoption, usage, and debugging. To address these challenges, we present LoongSage, an agent-native, fully asynchronous RL framework with a lightweight core of roughly 14K lines of code. Built upon Megatron for distributed training and SGLang for high-throughput inference, LoongSage centralizes orchestration within a single controller process and encapsulates all GPU-resident components as Ray actors. It introduces AgentFlow, a unified execution paradigm that orchestrates both white-box and black-box agents. For asynchronous execution, it integrates sliding-window strategies that bound trajectory staleness, and employs TransferMesh, a topology-aware synchronization mechanism that routes weights after each step over the fastest transport available to each receiver. Beyond agentic RL, LoongSage incorporates a memory-efficient Teacher Manager that decouples the scaling of teacher models from GPU memory constraints, natively supporting Multi-Teacher On-Policy Distillation. Extensive evaluations across four complete training runs, spanning complex reinforcement learning and multi-teacher distillation, demonstrate that task rewards improve steadily while the train-inference gap is kept under continuous control, and the fully asynchronous pipeline removes synchronous idle bubbles, sustaining hardware utilization.

cs.LG↗

arXiv2026

Infinitely many closed geodesic images on simply connected compact rank-one symmetric spaces

We prove that every smooth Riemannian metric on the underlying smooth manifold of a simply connected compact rank-one symmetric space, namely $S^n$ $(n\ge2)$, $\CP^m$ $(m\ge1)$, $\HP^m$ $(m\ge1)$, or $\OP^2$ admits infinitely many geometrically distinct prime closed geodesics. Here a prime closed geodesic is a nonconstant closed geodesic which is not an iterate of a shorter one, and geometric distinctness means that the images in the manifold are pairwise different. Thus iterating a geodesic, changing its starting point, or reversing its orientation does not count as a new geodesic. No nondegeneracy assumption on the metric is imposed. These manifolds have bounded loop-space Betti numbers, so the usual homological growth criterion does not yield this conclusion. We use the length filtration together with rotation and reversal of loops. A controlled cut construction gives a uniform filling estimate in the based loop space, with a fixed increase in length independent of degree, which we transfer to free loops. We then calculate global equivariant classes associated with odd and even iterates. Assuming finitely many prime geodesic images forces incompatible requirements at a class's first critical level. The proof treats spheres and projective spaces within a common framework and does not assume a sphere infinitude theorem.

math.DG↗

arXiv2026

An Adjoint Criterion for Left-Invariant $m$-Modified Conformal Vector Fields on Even-Dimensional Lie Groups

For a fixed positive real constant $m$, a real Lie algebra of even dimension $2k$ admits a nonzero left-invariant $m$-modified conformal vector field for some nondegenerate metric if and only if an adjoint operator has a nonzero real eigenvalue. In that case, a solution exists with neutral signature $(k,k)$ and zero conformal factor. As an application, we classify the admissible four-dimensional Lie algebras and metric signatures; the only Lorentzian exceptions among the admissible algebras are $\mathfrak{aff}(\mathbb{C})$ and the diamond algebra. On a pseudo-Riemannian manifold of arbitrary indefinite signature, a nonzero $m$-modified conformal vector field forces lightlike geodesic incompleteness, even when its conformal factor is variable.

math.DG↗

arXiv2026

HiFun: a Hierarchical Framework for Efficient Functional Dexterous Manipulation Learning

While multi-fingered dexterous robot hands have the potential to provide human-level dexterity, using them for functional manipulation remains difficult because task success depends on precise contact, coordinated arm-hand motion, and reactive recovery from execution errors. Precise contact depends on complex contact dynamics that are difficult to model accurately in simulation, motivating real-world learning. However, real-world demonstrations in the full arm-hand control space are costly and noisy because each trajectory must coordinate arm motion with high-DoF multi-finger contact. Small execution deviations can further cause slipped contact or premature force application, making recovery from off-nominal contact states important for robust execution. To deal with these challenges, we propose HiFun, a hierarchical real-world RL framework for functional dexterous manipulation tasks. Our key insight is to decouple the control complexity into a functional hand skill for precise contact execution and an arm-hand coordination policy for arm control and hand-skill activation. This design enables effective real-time human intervention for reactive recovery learning and efficient data collection with Dynamic Movement Primitive (DMP) augmentation. Across six real-world tasks on two dexterous hands, HiFun achieves 98.3% average success within one hour of online training and remains robust to dynamic disturbances. The project page is available at https://hly-123.github.io/HiFun/.

cs.RO↗

arXiv2026

Learning When to Trust Medical Knowledge: Regression-Safe Routing for Clinical Reasoning

Clinical reasoning LLMs need external medical knowledge, but retrieved evidence and reasoning chains can be outdated, poisoned, or hallucinated. Harmful knowledge can cause regression: changing a correct answer into an incorrect one. We construct a medical knowledge-poisoning benchmark covering retrieved evidence and reasoning chains, with seven attack families filtered by a three-judge "subtle-but-wrong" vote and paired clean exams that separate poisoning effects from question difficulty. No fixed knowledge-use strategy is robust across these settings: strong RAG methods trade robustness to poisoned evidence against accuracy when evidence is useful. We formulate knowledge use as per-instance routing and introduce Med-TrustRouter, a lightweight controller that keeps the solver frozen. From leakage-free runtime signals, it estimates each action's expected correctness gain and uncertainty, preserving the base answer unless another action is confidently better. Across seven medical reasoning settings, Med-TrustRouter achieves 78.9 macro and 69.2 worst-case accuracy, matching the strongest competitor on macro accuracy while substantially improving the worst-case floor. Its gains transfer across solvers and evidence judges, unlike a hand-coded router using the same signals. Similar performance across linear, tree, and neural action-value heads suggests that the benefit comes from routing rather than estimator complexity. These results support learning when to trust external knowledge for robust clinical reasoning.

cs.AI↗

arXiv2026

A diameter gap for irreducible compact symmetric spaces of fixed rank

We prove that, for every positive integer $r$, the positive normalized diameters of isometric quotients of simply connected irreducible compact Riemannian symmetric spaces of rank $r$ have a uniform positive lower bound. The bound is independent of the dimension and of the acting group, which need not be connected. For the higher-dimensional Grassmannian families, we obtain an explicit bound for positive-dimensional acting groups by constructing nonconstant invariant functions of bounded trigonometric degree. The real and complex cases use moment maps. The quaternionic case follows from a quartic rigidity theorem proved by means of invariant four-forms and the classification of compact quaternionic symmetric spaces. We also give examples showing why the rank and the relative scales of product factors must be controlled.

math.DG↗

arXiv2026

On the Orbital Evolution of Giant Planet Pair Embedded in a Gaseous Disk. III. simultaneous gas accretion and the final mass ratio

Recent exoplanet surveys reveal a systematic trend in the mass ratios of giant planet pairs ($> 30 M_\oplus$). Although the $M_{outer}/M_{inner}$ is broadly centered near unity, it exhibits a dependence on orbital separation. Compact pairs tend to have comparable masses, whereas widely separated pairs show outer-heavy configurations. To investigate the origin of this trend, we perform two-dimensional hydrodynamic simulations that couple gas accretion and orbital migration of two giant planet embryos with initial masses of $15-30 M_\oplus$. We explore the effects of disk viscosity, surface density slope, and initial mass ratio on the coupled evolution of planetary growth and migration. We find that convergent migration drives the planets to open a common gap, restructuring the gas flow. This configuration induces asymmetric accretion, in which the inner planet becomes starved as gap clearing suppresses its gas supply, while the outer planet continues to accrete from the high-density region at the outer gap edge. Consequently, the system naturally evolves toward mass-ratio convergence, and in many cases produces an inversion of the initial core-mass hierarchy. We further show that migration outcomes are highly sensitive to initial conditions. Systems with initially more massive inner cores tend to diverge, whereas steeper surface density profiles and more massive outer cores favor convergent migration and subsequent resonance capture. Overall, our results identify common-gap-driven asymmetric accretion as a unified mechanism linking orbital architecture, resonant capture, and the observed mass-ratio demographics of giant planet pairs.

astro-ph.EP↗

arXiv2026

Positive sectional curvature on homotopy nine-spheres

We determine which homotopy nine-spheres admit positive sectional curvature: they are precisely the four smooth types with vanishing real Dirac index. The other four admit no metric of nonnegative sectional curvature. In particular, we construct a strictly positively curved metric on the exotic Kervaire nine-sphere, answering the longstanding nonnegative-curvature existence question in this dimension. The proof keeps explicit attaching maps throughout. A quaternionic commutator realizes the Kervaire sphere, a mixed representation realizes the Hopf image of the exotic eight-sphere, and a four-periodic normal-framing formula identifies the remaining index-zero class. Following the compatible-disk method, we give the curvature and boundary estimates for the first two models, including an explicit lower bound for the Kervaire metric in terms of the curvature and first covariant-derivative norms of a fixed connection, with the base scale and fiber normalization fixed. For the third model, we construct a positive metric on a cylinder joining a quotient metric to its pullback by a marked right Hopf twist. A four-form modification controls the connection terms involving both left and right quaternionic multiplication. Two positive disk fillings close the cylinder with the prescribed attaching map. A local convolution lemma supplies the smooth gluing while preserving a strict sectional-curvature lower bound.

math.DG↗

arXiv2026

On the Structure of Sub-Riemannian Geodesic Orbit Manifolds

We study homogeneous sub-Riemannian manifolds whose normal extremals are orbits of one-parameter subgroups of the acting group. Our first main result shows that every sub-Riemannian geodesic orbit manifold satisfies the Goh condition and consequently admits no strictly abnormal length minimizers. Our second main result classifies all compact, connected, simply connected sub-Riemannian geodesic orbit manifolds with nonabelian simple isotropy group, thereby extending the Riemannian classification of Chen, Nikolayevsky and Nikonorov.

math.DG↗

arXiv2026

SLDR: Defending Against Malicious Fine-tuning via Selective Layers Recovery and Dynamic Routing

Fine-tuning-as-a-service enables users to adapt aligned large language models (LLMs) to specialized tasks, but malicious fine-tuning can erode refusal behavior while preserving task performance on legitimate inputs. We revisit recent layer-wise safety diagnostics and find that safety sensitivity is signed: scaling different layers can strengthen refusal, weaken it, or have little effect. Motivated by this observation, we propose SLDR, a post-fine-tuning defense based on Selective Layers Recovery and Dynamic Routing. SLDR trains a LoRA recovery adapter only on the layers with the maximum and minimum sensitivity scores in the signed spectrum, and uses representation-based dynamic routing inference to activate the adapter only for malicious queries. Across four model architectures, five downstream tasks, and four harmful benchmarks, SLDR substantially reduces harmful outputs while preserving downstream utility. On Llama3.1/SST2, SLDR reduces the average harmful score from 11.54 to 0.08 while maintaining downstream accuracy, and the harmful score remains near zero under poisoning ratios up to 0.9. The code is available at https://github.com/Stardust457/SLDR.

cs.CR↗

arXiv2026

Agentic RCA for Internet-Scale Services Using Constrained Creativity

System administrators of Internet-scale services need to resolve failure incidents to maintain reliability of such services. Ideally, we want a troubleshooting system to be: (1) expressive to known and unknown incidents with high accuracy; (2) cost efficient at scale; (3) explainable to provide actionable insights operators can act on; and (4) entail low effort from the operators. Unfortunately, most existing systems, including emerging LLM-assisted agentic workflows and structured frameworks for authoring diverse RCA algorithms fall short of achieving all four requirements. We present E4, a novel agentic system for troubleshooting for Internet-scale services. E4 embodies the paradigm of constrained creativity that combines the best of LLM-assisted automation and exploration with the explainability and efficiency of a structured approach. Instead of allowing an LLM agent to write arbitrary code or generate arbitrary responses, we provide the agent a restricted DSL to generate its response via simple loop-free data flow programs. This DSL, equipped with high level operators for troubleshooting, makes E4's output accurate, verifiable and explainable. On a mix of synthetic and real-world workloads, E4 achieves up to 62% better accuracy compared to state-of-the-art solutions, while providing more explainable responses at up to 12x reduced cost.

cs.NI↗

arXiv2026

APHABAMAS: An analytical phantom-based scheme for assessing the accuracy of high-resolution 3D MRI motion-artifact simulations

Purpose: Motion compromises the utility of high-resolution 3D MRI, an established tool in quantitative neuroimaging research. Deep learning-based methods have shown promise for mitigating motion-induced artifacts, but their development typically requires simulated motion-corrupted data. Several open-source tools exist for this task, each implementing different algorithms. However, no scheme currently exists for evaluating the accuracy of these simulations, making it difficult for users to choose the most suitable tool. This study aims to develop such a scheme. Methods: The essential ingredient of the desired scheme is a ground-truth reference that does not suffer from sampling-induced error. To meet this requirement, the proposed scheme, APHABAMAS, leverages a digital phantom whose image- and Fourier-domain representations can be expressed analytically under arbitrary rigid-body transformations. APHABAMAS is used to quantify sampling-induced errors of three existing algorithms under synthetic and tracking-data-derived motion trajectories at different motion severity levels. Results: All three algorithms produce distinct but visually plausible artifacts, demonstrating the need for a ground-truth reference. APHABAMAS shows that the algorithm utilizing the uniform-to-non-uniform (Type-2) NUFFT provides the most consistent agreement with the ground-truth reference across motion types and severity levels. The non-uniform-to-uniform (Type-1) NUFFT-based algorithm, which does not mimic the MR acquisition process, deteriorates substantially with increasing motion severity. Conclusions: APHABAMAS provides a rigorous tool for assessing the accuracy of high-resolution 3D MRI motion-artifact simulations. It enables accuracy-based assessment of three existing algorithms, thereby facilitating informed selection of the most suitable one for synthesizing motion-corrupted data.

physics.med-ph↗

arXiv2026

Resolving quantitative MRI model degeneracy in self-supervised machine learning

Quantitative MRI (qMRI) estimates tissue properties of interest from measured MRI signals. This process is conventionally achieved by model fitting, whose computational expense limits qMRI's clinical use, motivating recent development of machine learning-based methods. Self-supervised approaches are particularly popular as they avoid the pitfall of distributional shift that affects supervised methods. However, it is unknown how such methods behave if similar signals can result from multiple tissue properties, a common challenge known as model degeneracy. Understanding this is crucial for ascertaining the scope within which self-supervised approaches may be applied. To this end, this work makes two contributions. First, we demonstrate that model degeneracy compromises self-supervised approaches, motivating the development of mitigation strategies. Second, we propose a mitigation strategy based on applying appropriate constraining transforms on the output of the bottleneck layer of the autoencoder network typically employed in self-supervised approaches. We illustrate both contributions using the estimation of proton density fat fraction and $R_2^*$ from chemical shift-encoded MRI, an ideal exemplar due to its exhibition of degeneracy across the full parameter space. The results from both simulation and $\textit{in vivo}$ experiments demonstrate that the proposed strategy helps resolve model degeneracy.

physics.med-ph↗

arXiv2026

TRACER: Texture-Robust Affordance Chain-of-Thought for Deformable-Object Region Grounding

The central challenge in robotic manipulation of deformable objects lies in aligning high-level semantic instructions with physical interaction points under complex appearance and texture variations. Existing vision-based affordance prediction methods often suffer from boundary overflow and fragmented functional regions. To address these issues, we propose TRACER, a Texture-Robust Affordance Chain-of-Thought for Deformable-Object Region Grounding framework that maps hierarchical semantic reasoning to appearance-robust and physically consistent interaction regions. Specifically, a Tree-structured Affordance Chain-of-Thought (TA-CoT) decomposes high-level task intentions into hierarchical affordance-semantic instructions. A Spatially-Constrained Boundary Refinement (SCBR) mechanism suppresses prediction spillover and guides responses toward valid object regions. Furthermore, an Interactive Convergent Refinement Flow (ICRF) refines dispersed affordance responses into coherent regions, improving spatial continuity and physical plausibility. Experiments on the Fine-AGDDO15 dataset and a real-world robotic platform demonstrate that TRACER improves affordance grounding precision across diverse textures and patterns. It also enhances long-horizon manipulation success, bridging high-level semantic reasoning and low-level physical execution. The source code and dataset will be made publicly available at https://github.com/Dikay1/TRACER.

cs.RO↗

arXiv2026

VIDiff: Translating Videos via Multi-Modal Instructions with Diffusion Models

Diffusion models have achieved significant success in image and video generation. This motivates a growing interest in video editing tasks, where videos are edited according to provided text descriptions. However, most existing approaches only focus on video editing for short clips and rely on time-consuming tuning or inference. We are the first to propose Video Instruction Diffusion (VIDiff), a unified foundation model designed for a wide range of video tasks. These tasks encompass both understanding tasks (such as language-guided video object segmentation) and generative tasks (video editing and enhancement). Our model can edit and translate the desired results within seconds based on user instructions. Moreover, we design an iterative auto-regressive method to ensure consistency in editing and enhancing long videos. We provide convincing generative results for diverse input videos and written instructions, both qualitatively and quantitatively. More examples can be found at our website https://ChenHsing.github.io/VIDiff.

cs.CV↗

arXiv2026

From bits to atoms: A survey of cross-layer safety in Embodied AI

Embodied Artificial Intelligence (AI) introduces cyber-physical risks because digital failures and attacks can cause physical harm. This survey organizes embodied AI safety through a four-layer reference architecture -- comprising the physical interaction, system and middleware, algorithm and decision, and cloud-edge and data layers -- and uses it to connect vulnerabilities, propagation paths, and safeguards across the system stack. It traces the evolution of safety paradigms and examines how faults and attacks cross architectural boundaries to produce physical consequences. Furthermore, the survey systematizes composite attack chains, outlines multi-layered defense strategies, and reviews representative embodied safety benchmarks. Ultimately, we advocate for unifying software alignment with rigorous hardware constraints to build trustworthy systems capable of operating safely within human-centric environments.

cs.RO↗

arXiv2026

When History Misleads: Asymmetric Margin Supervision for Instruction-Guided LLM Generative Recommendation

In instruction-guided generative recommendation, LLM-based recommenders need to balance two goals: responding to the user's current request and aligning with the preferences in their interaction history. When the two conflict, history events can override the request. We show that turning the effect of individual history events into supervision faces two obstacles. First, the events that most influence a recommendation are not necessarily the ones that support the target item. Second, removing a misleading event can raise the target's score but a competing item's score even more, so a higher target score alone does not guarantee a better ranking. We propose Asymmetric Intervention-Guided Margin Supervision (AIMS), which converts the effect of removing individual history events into ranking supervision. For training requests already ranked correctly, a frozen reference model identifies request-specific deletions that improve both the target's score and its margin over a competitor near the recommendation cutoff. These margins serve as training targets, while the complete history is retained as input. Training combines cross-entropy with an asymmetric auxiliary loss that penalizes margin shortfalls and routes its gradient only through the competitor score. Inference is unchanged, requiring no history editing or deletion search. Across six LLM backbones on an industrial dataset and two public benchmarks, AIMS improves Recall and NDCG over strong baselines. Ablations support request-specific margins and asymmetric supervision, and the selected deletions preferentially remove constraint-violating history.

cs.IR↗