arXiv subjects
Publications and source records attributed to Zhi Li.
In this paper, we establish a logarithmic Bogomolov--Sommese vanishing theorem in terms of numerical dimension for pseudo-effective line bundles on compact K\"ahler manifolds. As an application, we obtain a rigidity result with vanishing second Chern class for logarithmic cotangent bundles by combining the vanishing theorem with a structure theorem of Iwai and Matsumura.
Bridging model-based control and learned policies in long-horizon manipulation has harbored a silent disagreement: control executes specified objectives, learning amortizes that behavior into a reactive policy, yet existing protocols discard task semantics, leaving rewards hand-crafted and behavior drifting from what control verified.We introduce Semantically UNified (SUN) Programs, typed executables where geometric and contact relations are defined once and compiled into aligned Model Predictive Control (MPC) costs, satisfaction predicates, RL rewards, transition guards, and diagnostics. Our system, Kuafu, driven by large vision language systems, automatically synthesizes SUN Programs from language and scene semantics, screens feasibility via MPC, and retains semantics while training stage-conditioned policies. Across nine tasks, Kuafu achieves 82.03% macro-success, outperforming sparse-reward (35.67%) and Stage-BC (24.75%) baselines. At 8192-way scale, it generates 10.57x the successful trajectory time per hour of human teleoperation. With 500 trajectories per task, Kuafu data trains DP3 policies to 46.0% simulation success (vs. 22.4% for alternatives) and 34.7% on physical Franka and Kinova robots. These results establish that simulation-screened task semantics can effectively amortize control into robust policies, without demonstrations or manual dense rewards, unifying symbolic planning and data-driven execution.
We study the multifractal criticality in a generalized Su-Schrieffer-Heeger model. The results show that the system supports not only critical phases but also re-entrance multifractalization (REM). By mapping the hopping term to an effective potential, we analytically prove that although the model has no explicit unbounded potential, a hidden unbounded potential is actually present-this is the key mechanism driving the emergence of multifractal critical phases. Moreover, one can get a condition where the competition between the explicit and hidden unbounded potentials is exactly balanced. Under this condition, the multifractal critical phase vanish, and the system returns to the extended phase. Based on this mechanism, we achieve both demultifractalization and re-entrant multifractalization. Finally, we double check the theoretical predictions through wave packet dynamics, and the numerical results are consistent with our theoretical analysis. This work broadens our understanding of how unbounded potentials induce multifractal critical phases, providing a theoretical basis for designing new systems with multifractal critical phases.
In clinical motor imagery brain-computer interface (MI-BCI) decoding, cross-day transferability and online operation remain two critical challenges. Hypergraphs can improve transferability by capturing higher-order sample relationships, yet existing hypergraph-based methods for online emotion recognition neglect the cross-day benefits of Riemannian geometry widely adopted in EEG transfer learning. To bridge this gap, we propose the Multi-feature Riemannian Hypergraph (MRieHy), a framework tailored for online test-time adaptation in MI-BCI decoding that leverages Riemannian geometry to strengthen cross-day transferability. MRieHy first computes Riemannian means of covariance matrices from cross-day training data to align multi-day distributions. It then constructs a hypergraph over covariance matrices using Riemannian distance, complemented by a second hypergraph over deep features built with cosine similarity. The two hypergraphs are fused via adaptively learned combination weights, jointly optimized with the label projection matrices. During online testing, MRieHy maintains a first-in-first-out buffer of recent samples, performs Riemannian alignment on the buffered data, and decodes with the learned hypergraph. Extensive experiments on a private four-class ECoG dataset and two public four-class EEG datasets validate that MRieHy achieves notable performance gains over state-of-the-art baselines.
Visual loco-manipulation policies that can generalize to novel scenes and objects have long been a goal of robotics research. However, today's data-hungry algorithms make collecting sufficient demonstrations a struggle for tabletop manipulation, and even more so for humanoids that must also walk and balance. Learning from simulated data and transferring that behavior to the real world, as is commonly done in locomotion, sidesteps this struggle, so we replicate that recipe for loco-manipulation. In doing so, we find that cloning synthetic demonstrations results in a low performance ceiling no matter the amount of training data. Reinforcement learning breaks through it, and refining the cloned policy with Flow-GRPO on a single sparse reward yields performance that synthetic behavior cloning cannot match. Together, these stages form our end-to-end sim-to-real pipeline spanning more than 150,000 scenes, which we use to train FetchMan. We evaluate it on FetchMan-Bench, a simulation benchmark we release, and deploy it zero-shot on a real Unitree G1, where our single-object reach-and-pick policy walks to and grasps a target across unseen scenes at 73.3% success. Finally, we extend this recipe to multi-object training, a first step toward loco-manipulation generalist policies at this data scale.
We present a deep photometric analysis of the main-sequence (MS) population in the old, metal-rich open cluster (OC) NGC 6791 using Gaia Data Release 3 data. After correcting for differential reddening, we use the Bayesian model comparison to test whether stellar rotation can account for the observed MS broadening and find that a rotation-dominated interpretation is strongly disfavored. We therefore infer that unresolved multiplicity is the primary contributor to the photometric offsets. We derive a high-q companion fraction of $54.3\% \pm 2.8\%$ for systems with $q \gtrsim 0.5$, significantly higher than typical values reported for most OCs and the field. The inferred offset distribution is not consistent with a flat mass-ratio distribution but instead shows an excess toward high mass ratios ($q \sim 0.8$--$1.0$), suggestive of preferential survival of hard binaries in a dynamically evolved environment. We also identify a population of stars lying above the equal-mass binary limit ($\Delta G > 0.75$ mag), which is difficult to explain with ordinary MS binaries alone and is plausibly interpreted as candidate unresolved triple or higher-order multiple systems. A Kolmogorov--Smirnov test, together with Monte Carlo label-shuffling experiments, shows no statistically significant difference between the projected radial distributions of the single-star and binary/multiple populations within the observed field. Taken together, these results are consistent with the picture that NGC 6791 is the dynamically processed inner remnant of a once more massive cluster.
There are 56,804 public agent skills today, and teams write many more privately. The dominant delivery model is installation: once installed, a skill's description remains in the system prompt, competing for fewer than 100 reliable trigger slots. This leaves the long tail with no practical path to use and forces teams' own playbooks to compete for the same scarce space. We observe that installation bundles three separable functions: content, persistence, and automatic triggering. Only the last requires prompt residency. We therefore propose @skills, an open protocol that separates them. A path addresses any skill, subtree, or collection, and reading a skill is sufficient to use it, so nothing is installed or made resident. The operation vendors a copy at the same path into a project's Git-tracked tree for adaptation and ownership. The operation adds one .gitignore-style line, the only element that costs prompt residency. A directory is a menu, making bundles ordinary directories rather than all-or-nothing units. The protocol requires no manifest, lockfile, or registration, and SKILL.md remains unchanged. @skills is additive, ships as an installable package, and turns any agent that can read files and run commands into a client through a single instruction file. Its open specification is at https://github.com/SylphAI-Inc/atskills and it is implemented in the AdaL CLI at https://adalagent.ai . Because paths address skills well but cannot find them, the protocol is paired with a free hub at https://atskills.one for corpus-wide search and ranking, repository-free hosting, private and team collections, and one-screen authoring. The hub is optional: gh: and local paths resolve without it, and indexed GitHub skills retain their gh: identities. Install less, use more.
Delegating a web task involves more than asking a question; it requires transferring a policy: what to verify, how to handle uncertainty, which preferences matter, and when to stop. Yet, current live-web agents are evaluated solely on the final answer, ignoring the policy constraints that define the delegation. A plausible final answer can conceal violations of that policy. Our full live audit reveals this critical gap: a strong controller completes 99.2% of tasks but honors all policy constraints in only 38.8% of cases. Finishing does not imply fidelity. WebRider bridges this gap by formalizing the delegated policy as an intent contract---an operational record of goals, constraints, evidence obligations, answer form, and task-local persona controls that must hold even as web pages change. WebRider employs a hierarchical architecture: a top-layer controller maintains the contract, a middle layer realizes intentions as guarded executable actions, and a tool layer executes these actions via browser, search, and maps tools. Our benchmark, RiderBench, evaluates this design on 4,096 live-web contracts across 42 public websites, auditing both the internal contract state and the visible user experience to determine if a rollout preserved its policy and if the steps were persona-consistent. The guarded middle interface also serves as a high-quality training signal; an 8B action-policy model trained through this interface outperforms executable-only baselines under a fixed controller. By making the browsing path a first-class object, WebRider enables a system that is auditable, human-judgeable, and learnable without conflating action realization with final-answer decisions. Dataset URL: hf.co/datasets/WebRider/WebRider.
Image-goal navigation is a fundamental capability for embodied AI, yet its practical deployment is strained by strong prior assumptions. Existing methods predominantly rely on temporally ordered video streams or auxiliary sensors (e.g., depth, LiDAR) to maintain spatial consistency. These sequential and multimodal dependencies severely restrict scalability, especially when deploying robots using crowd-sourced or pre-recorded unordered image collections. When temporal priors are removed, current methods struggle with severe perceptual aliasing, noisy associations, and catastrophic mapping failures. To address this underexplored challenge, we propose Unordered Landmark Visual Navigation (ULVN), a unified RGB-only framework free from temporal and odometric priors. ULVN systematically mitigates error accumulation by integrating mapping, localization, and planning. Specifically, it constructs a robust 2D topological map directly from unstructured images via calibrated geometric verification and maximum spanning forest refinement. For closed-loop execution, ULVN abandons sequential heuristics, utilizing a graph-based belief propagation filter with entropy-adaptive fusion for global localization and dynamic subgoal planning. Extensive experiments in simulation and real-world deployments demonstrate that ULVN significantly outperforms state-of-the-art methods.
Registration between cone-beam computed tomography (CBCT) and intraoral scans (IOS) is essential for patient-specific surgical planning. However, disparate imaging modalities, limited overlap, and large pose offsets make automated registration unreliable. Consequently, clinical registration remains dependent on conventional geometry pipelines and manual clinician adjustment. To address these challenges, we propose APCReg, an anatomical-prior-guided coarse-to-fine framework for global registration and reliability-controlled residual correction. Specifically, multi-view anatomical coarse registration (MACR) performs ordered orthogonal projection alignment (buccal, proximal, and occlusal) to decompose the six-degree-of-freedom search before three-dimensional refinement. Overlap-aware residual registration (OARR) combines shared KPConv features, a folded arch-length cue, overlap-gated cross-attention, and Sinkhorn matching. Finally, dental-arch-structured hypothesis selection evaluates diverse poses on held-out reliable correspondences, while a ground-truth-free coarse-retention guard conditionally retains a geometrically reliable coarse pose. On 60 held-out jaw pairs, APCReg achieves a submillimeter mean Chamfer distance of 0.87 mm and a Hausdorff distance of 2.92 mm under this evaluation protocol, and ranks first across the six reported metrics among the evaluated open-source baselines.
To probe a bipartite quantum system, one may use arbitrary global operators or restrict to product operators acting separately on the two subsystems. We determine the sharp universal comparison between the resulting norms. For every $z\in M_n\otimes M_m$, we prove $\|z\|_1\leq\sqrt{2}\min\{n,m\}\|z\|_\varepsilon$, where $\|\cdot\|_1$ is the trace norm and $\|\cdot\|_\varepsilon$ is the injective tensor norm associated with the trace norms on $M_n$ and $M_m$. To prove the upper bound, we establish an $L_1$ noncommutative Khintchine inequality whose random coefficients are the entries of a Haar unitary. We also show that the coefficient $\sqrt{2}$ is sharp. As applications, we show that the same sharp constant governs the gap between bipartite correlation measured in trace norm and that measured by a correlation function, and obtain an improved universal upper bound for quantum data hiding. The upper bound has also been formalized and machine-checked in Lean.
Quantum data hiding asks how much distinguishing power can be lost when global measurements are restricted to local measurements and classical communication. In this work, we establish sharp results and improved bounds for several natural classes of restricted measurements. For bipartite systems on $\mathbb C^n\otimes\mathbb C^m$, we prove that the optimal data-hiding ratios against separable and LOCC measurements are both $\min\{n,m\}$. This result follows from a stronger result that, for every $2\le p\le\infty$, the largest centered Schatten $p$-ball whose associated binary measurements are implementable by finite-round LOCC has radius $\min\{n,m\}^{2/p-1}$. This strengthens the classic separable-ball theorems, while also providing an explicit finite-round LOCC implementation. For Alice-first one-way LOCC with Alice's local dimension equal to $n$, we prove that the optimal ratio is $(1+o(1))n$, with the upper bound obtained from a Gaussian rank-one POVM. For local operations without communication, we improve the universal upper bound to $(\pi\sqrt3/4+o(1))\min\{n,m\}$.
Coding-agent benchmarks increasingly cover long-horizon, end-to-end, and interactive development, but typically retain one requested outcome or a fixed change sequence. Sequential policies can process a pull-request (PR) queue one candidate at a time, but when queued PRs interact, maximizing safe delivery can require jointly deciding which changes to merge and in what order. We introduce BulkPR-Bench, an executable benchmark in which an agent must recover consequential PR relations and return a large safe subset in executable order under a rolling-release protocol. The suite contains 581 newly authored candidate PRs on frozen snapshots of 18 real repositories. Registered state-by-state repository execution, including hidden safety checks, validates the gold relation graph; an exact oracle then computes the largest safe subset. Our primary metric, Relational Delivery Score (RDS), scores safe delivery and correct rejection over relation groups from the realized merge trace; Global Safety-Gated Yield (Global-SGY) separately measures strict delivery of the realized whole-queue plan. Under the buffered primary protocol with batch size $K=32$, the three highest RDS estimates among the six models are 66.6%, 62.0%, and 57.9%, compared with 53.1% for the strongest sequential baseline. Only 8 of 324 model runs complete a queue exactly. Critical-relation recall ranges from 35.2% to 57.7%, and diagnostic runs supplied with the gold relations show substantial remaining headroom. Gains on relation groups therefore do not yet translate into dependable whole-queue governance.
The rapid adoption of artificial intelligence (AI)-driven workflows is transforming high-consequence government and enterprise systems into language-based, tool-using and increasingly autonomous infrastructures. While these workflows can delegate planning autonomously, security-critical execution should be strictly mediated. Conventional identity management services authenticate who may invoke a primitive, but remain agnostic to which workflow steps are authorized at runtime. An AI-driven workflow can still be hijacked by injection attacks into executing malicious actions that satisfy identity checks yet violate user intent. We propose Neural Cryptographic Services (NCS), a neuro-symbolic security enforcement plane interposed between neural planners and privileged tools. NCS decouples cognitive planning from execution authority: an untrusted neural planner drafts structured plans, while a deterministic symbolic controller gates execution using an offline-signed, hash-chained instruction stream. Specifically, NCS validates cryptographic signatures and hash chains incrementally, releasing a single instruction template at a time, and admitting a tool call only when its proposed parameters satisfy the constraints of the signed template. Out-of-order or altered tool calls fail-closed, and state transitions are logged for post-hoc auditing. NCS does not attempt to prevent neural planner compromise under injection; it guarantees that a compromised planner cannot dispatch actions outside the authorization. We evaluate NCS using AgentDojo, a custom argument-hijacking dataset, adaptive adversarial instructions, and TheAgentCompany. NCS drives attack success rates to near zero while preserving acceptable utility on benign workflows.
This paper presents a generative artificial intelligence (Gen AI) approach for forecasting, at a day-ahead stage, the largest single infeed (LSI) and largest single outfeed (LSO) on the Irish power system to assist in reserve dimensioning. Developed collaboratively between EirGrid, the electric transmission system operator (TSO) for Ireland, and GridZero.ai using the GridZero.ai platform, the system delivers accurate forecasts up to 38 hours ahead of real-time using limited data available before the day-ahead and intra-day energy market gate closure timings. Initial performance demonstrates an accuracy with a mean absolute percentage error (MAPE) that is only 1.1\% higher than the results possible using full market data (8-hours ahead). Thus, if this approach is integrated into operational systems and such high levels of accuracy are maintained, reserve procurement costs could be significantly reduced. The results also demonstrate the practicality and extensibility of AI-powered resource planning for TSOs.
Filesystem isolation in container ecosystems is often weakened by cross-boundary path misresolution, causing path traversal (PaTra) vulnerabilities. These vulnerabilities stem from insecure host-container interactions and have become increasingly pervasive as cloud systems mount shared resources, such as GPUs and agent workspaces, into containers to support AI workloads. Existing defenses remain inadequate. Kernel-level protections are intrusive, can destabilize system calls, and have therefore not been accepted into the Linux mainline. Detection methods rely on static rule matching or manual code auditing. Static rules can flag path-related functions but fail to capture the semantics needed to determine whether a host-container interaction exists, causing many false positives. Manual review requires domain expertise, making it costly, inefficient, and difficult to scale. To address this threat, we present Bulkhead, an automated framework that integrates large language models (LLMs) with formal methods for semantic vulnerability discovery and remediation. Bulkhead uses a multi-agent system to identify and repair PaTra vulnerabilities through multi-dimensional knowledge patterns generalized from known cases. It first applies high-risk functional patterns to locate entry points for cross-boundary interactions in containerized code, then uses call-chain patterns to recover the corresponding execution paths at suitable depth. The Detection pipeline analyzes these call chains against the application scenarios and threat model, identifying vulnerabilities such as missing security checks and TOCTOU flaws in cross-boundary interactions, and generating proof-of-concept (PoC) exploits for validation. These PoCs then guide patch generation. To ensure remediation correctness, the Patch pipeline performs assertion-driven verification using predefined model-checking templates.
Predicting where floodwater goes and how deep it gets, at high resolution and across large domains, remains computationally expensive with conventional hydraulic solvers, while purely data-driven surrogates are fast but lack physical guarantees and generalize poorly beyond their training events. We present Inunda, a GPU-native flood inundation model that solves the two-dimensional shallow water equations. Inunda uses a mass-conservative local-inertial scheme and runs multi-day events over millions of cells in minutes on a single GPU. Because every operator is autograd-compatible, the solver is differentiable by construction: model parameters can be estimated by gradient descent against gage observations through reverse-mode automatic differentiation of the full simulation. We demonstrate Inunda on three case studies. For a hindcast of Hurricane Harvey (2017) in Harris County, Texas, Inunda matches surveyed high-water marks to a mean absolute error of 0.67 m, competitive with or better than a suite of established flood models, and reaches a median gage water-level Nash Sutcliffe efficiency of +0.72, more than double the +0.31 of the operational National Water Model v3.0. For the July 2025 Central Texas flash flood, Inunda is driven by an 18-member 1-km convection-allowing precipitation ensemble to produce probabilistic flood forecasts whose skill improves systematically as lead time to the crest shortens. For a post-fire flash-flood application in the Rio Ruidoso burn scar, differentiable calibration recovers the saturated hydraulic conductivity as a spatially explicit field at the model's own resolution and traces its multi-year post-fire recovery. Inunda provides an open, end-to-end pipeline for real-event flood modeling that couples the accuracy of physics-based hydraulics with the calibration and coupling advantages of modern differentiable programming.
Quantum secret sharing is one of the core security technologies in the field of quantum communication. Aiming at the problems of high quantum resource consumption and the difficulty in balancing security and efficiency existing in current quantum secret sharing schemes under non-maximal quantum access structure(QAS), this paper focuses on the judgment criteria of non-maximal QAS and the construction methods of efficient and perfect quantum secret sharing schemes on a kind of non-maximal QAS, i.e. hyperstar with three hyperedges. Firstly, we characterizes the forbidden sets and intermediate sets within this non-maximal QAS, and provides a characterization for non-maximal QAS realizable by pure-state encoding. Secondly, the representative element QAS of hyperstar with three hyperedges is defined, which serves as a substructure of this hyperstar QAS, and belongs to non-maximal QAS. Furthermore, we propose a universal, efficient, and perfect quantum secret sharing scheme based on this hyperstar QAS using corresponding classical secret sharing scheme as an auxiliary. By deploying lightweight quantum resources to representative element access structures, the proposed scheme reduces the difficulty in quantum state preparation and distribution, and enhances its security and resource utilization efficiency.