arXiv subjects

Zhengyu Liu

Publications and source records attributed to Zhengyu Liu.

15 recordsLinked to original sources

arXiv2026

Same Request, Different Boundary: Evaluating Cybersecurity Assistance across Conversational Contexts

Large Language Models (LLMs) can solve complex problems, but their misuse in high-risk domains can lead to severe consequences. Model providers therefore restrict assistance for potentially harmful requests. Refusing all cybersecurity requests would therefore harm legitimate users. Providers need a mechanism to block malicious use without denying legitimate assistance to defenders. Existing cybersecurity-specific datasets evaluate this mechanism, but none considers the conversational context of a request. We introduce 3R-Bench (Refusal, Repetition, and Revision), a benchmark of 150 real-world cybersecurity requests augmented with two adversarial conversational settings, and evaluate eight LLMs on it. Prior assistant behavior strongly changes responses to an unchanged request: among 376 available pairs from a 400-pair panel, compliance rises from 62.0% after refused history to 85.1% after accepted history. The opposite pattern appears under dialogue decomposition. In comparison, compliance falls from 501/800 direct responses to 172/800 after dialogue; among 738 pairs returning model-authored text in both conditions, the decrease is 45.1 points. Failure feedback recovers only a small fraction of this loss.

cs.AI↗

arXiv2026

SoK: When Safe Agents Fail Together: The Security of Multi Agent LLM Systems

Safe agents can fail together. Multi-agent LLM systems (MAS) move information, state, decisions, and authority across principal boundaries, creating failures that local checks may miss. Without an execution-level view, a multi-agent setting can easily be mistaken for evidence of a genuinely multi-agent security effect. We thus systematize MAS security through an execution-centered analysis of 197 works, covering six interaction interfaces, four adversary positions, seven system-level risks, and eight recurring attack paths. We introduce an A-I-R framework that organizes attacks by adversary position, interaction interface, and resulting system-level risk, unifying otherwise fragmented attack mechanisms across MAS. We organize defenses through a five-part contract covering path target, observation, intervention, trust boundary, and recovery, and identify path closure and recovery as key challenges. We audit 44 evaluation and benchmark works and identify open challenges in isolating interaction effects, designing comparable and diagnostic metrics, supporting reuse across MAS designs, and evaluating open-system operation. Together, these findings motivate an interaction-aware view of MAS security: trace attacks end to end, test whether defenses close those paths, and evaluate system-level effects with appropriate counterfactuals.

cs.CR↗

arXiv2026

Revealing low-energy surfaces of multinary compounds by controlling surface coordination environments

When modeling surfaces of multinary compounds, conventional cleavage planes often cut through strongly bonded polyhedra, resulting in unphysical surface energies. Here, we introduce SALAMI (Symmetric Atomic Layers for Arbitrary Multinary Interfaces), a Python package that generates symmetric, charge-neutral, dipole-free, and low-energy slab models for multinary compounds. SALAMI performs combinatorial searches to selectively remove surface atoms and generate corrugated terminations that preserve optimal coordination environments. We applied this workflow to all symmetrically inequivalent crystallographic orientations with Miller indices up to 2 for two prototypical structures: the solid-state electrolyte Li3PS4 and the transparent conducting oxide ZnSb2O6. Density functional theory calculations reveal that Li3PS4 must preserve all PS4 units to achieve the minimum surface energy. For ZnSb2O6, low-energy surfaces are achieved by partial undercoordination of surface Sb atoms to SbO5 or SbO4 from the bulk SbO6, depending on the surface orientations. Compared to surface models generated with unconstrained coordination, applying constraints to achieve optimal local coordination environments significantly lowers surface energies, shrinking the volume of the predicted Wulff shape by approximately 20%. Our results demonstrate that meticulous control of local coordination environments is necessary for accurately predicting the surface energetics of multinary compounds.

cond-mat.mtrl-sci↗

arXiv2026

Optimizing Frequent Checkpointing via Low-Cost Differential for Distributed Training Systems

Distributed training of large deep-learning models often leads to failures, so checkpointing is commonly employed for recovery. State-of-the-art studies focus on frequent checkpointing for fast recovery from failures. However, frequent checkpointing generates numerous checkpoints, incurring substantial costs and thus degrading training performance. Recently, differential checkpointing has been proposed to reduce costs, but it is limited to recommendation systems, so its application to general distributed training systems remains unexplored. In this paper, we find that gradients generated during distributed training can be reused to construct differential checkpoints, while the former's size is smaller than the latter's, motivating us to reuse gradients for low-cost differential checkpointing. Based on this main idea, we propose \sysname, a frequent checkpointing framework for compression-enabled training systems that reuses compressed gradients as differential checkpoints, eliminating redundant differential computation and reducing checkpoint transmission cost. Furthermore, we extend gradient reuse to scenarios without gradient compression and propose \sysnameplus, which employs layer-wise-reuse snapshotting and incremental-merging persistence to overlap checkpointing with training execution. Experiments on diverse workloads, including billion-parameter-scale models, demonstrate that \sysname and \sysnameplus significantly reduce checkpointing overhead and enable checkpointing at frequencies as high as once per iteration, reducing training time by up to 89.2\% and 81.2\%, respectively.

cs.DC↗

arXiv2026

Buzz to Boom: Detecting Message Progression Vulnerabilities in Electron Applications via Segmented Directed Fuzzing

Electron is a popular framework for building cross-platform desktop applications using web technologies. Such applications consist of multiple processes with different privilege levels that communicate via message passing. When inter-process messages carry attacker-controlled inputs, they can propagate across processes and reach privileged APIs, e.g., command execution. Such a message propagation behavior is characterized as Message Progression Vulnerabilities (MPVs). The exploitation of MPVs is challenging because it often requires multiple steps, e.g., first arbitrary code execution in one process via message passing, and then command injection in another process using another message crafted in the first process. To our knowledge, existing works on Electron security only study unsafe configurations and malicious Document Object Model (DOM) content, i.e., they cannot detect or exploit these vulnerabilities that need to be triggered by complex cross-process exploits via message passing. We present Proton, a segmented directed fuzzing framework for detecting MPVs. Our key insight is to decompose end-to-end fuzzing into per-process segments along message-passing boundaries, where the goals of fuzzing each segment are either: (i) reaching a sink in the current process or (ii) propagating the payload to the next process, to enable the exploration of another process. In the second case, the messages seed the corpus of the next segment. Finally, Proton synthesizes crash inputs from each process to validate end-to-end exploits. We evaluate Proton against 589 real-world Electron applications, resulting in 23 zero-day MPVs. Among them, 22 lead to OS command execution, including projects with over 50k GitHub stars. We responsibly disclosed all findings. To date, we have received 13 acknowledgments, 11 fixes, and 11 CVEs, including a bug bounty from Vercel.

cs.CR↗

arXiv2026

Comment and Control: Hijacking Agentic Workflows via Context-Grounded Evolution

Automation platforms such as GitHub Actions and n8n are increasingly adopting so-called agentic workflows, which integrate Large Language Model (LLM) agents for tasks such as code review and data synchronization. While bringing convenience for developers, this integration exposes a new risk: An adversary may control and craft certain inputs, such as GitHub issue comments, to manipulate the LLM agent for unwanted actions, such as credential exfiltration and arbitrary command execution. To our knowledge, no prior academic work has studied such a risk in agentic workflows. In this paper, we design the first detection and exploitation framework, called JAW, to hijack agentic workflows hosted on automation platforms via a novel approach called Context-Grounded Evolution. Our key idea is to evolve agentic workflow inputs under the contexts derived from hybrid program analysis for hijacking purposes. Specifically, JAW generates agentic workflow contexts through three analyses: (i) static path-feasibility analysis to identify feasible agent-invocation paths and the input constraints required to trigger them, (ii) dynamic prompt-provenance analysis to determine how that input is transformed and embedded into the LLM context, and (iii) capability analysis to identify the actions and restrictions available to the agent at runtime. Our evaluation of JAW on GitHub workflows and n8n templates showed that 4714 GitHub workflows and eight n8n templates can be successfully hijacked, for example, to leak user credentials. Our findings span 15 widely-used GitHub Actions, including official GitHub Actions for Claude Code, Gemini CLI, Qwen CLI, and Cursor CLI, and two official n8n nodes. We responsibly disclosed all findings to the affected vendors and received many acknowledgements, fixes, and bug bounties, notably from GitHub, Google, and Anthropic.

cs.CR↗

arXiv2024

A self-consistent Hartree theory for lattice-relaxed magic-angle twisted bilayer graphene

For twisted bilayer graphene close to magic angle, we show that the effects of lattice relaxation and the Hartree interaction both become simultaneously important. Including both effects in a continuum theory reveals a Lifshitz transition to a Fermi surface topology that supports both a ``heavy fermion" pocket and an ultraflat band ($\approx 8~{\rm meV}$) that is pinned to the Fermi energy for a large range of fillings. We provide analytical and numerical results to understand the narrow ``magic angle range" that supports this pinned ultraflat band and make predictions for its experimental observation. We believe that the bands presented here are accurate at high temperature and provide a good starting point to understand the myriad of complex behaviour observed in this system.

cond-mat.str-el↗

arXiv2024

Effects of Grain Boundaries and Surfaces on Electronic and Mechanical Properties of Solid Electrolytes

Extended defects, including exposed surfaces and grain boundaries, are critical to the properties of polycrystalline solid electrolytes in all-solid-state batteries (ASSBs). These defects can significantly alter the mechanical and electronic properties of solid electrolytes, with direct manifestations on the performance of ASSBs. Here, by building a library of 590 surfaces and grain boundaries of 11 relevant solid electrolytes $-$including halides, oxides, and sulfides$-$ their electronic, mechanical, and thermodynamic characteristics are linked to the functional properties of polycrystalline solid electrolytes. It is found that the energy required to mechanically ``separate'' grain boundaries can be significantly lower than in the bulk region of materials, which can trigger preferential cracking of solid electrolyte particles in the grain boundary regions. The brittleness of ceramic solid electrolytes, inferred from the predicted low fracture toughnesses at the grain boundaries, contributes to their cracking under local pressure imparted by Lithium or Sodium penetration in the grain boundaries. Extended defects of solid electrolytes introduce new electronic ``interfacial'' states within bandgaps of solid electrolytes. These interfacial states alter and possibly increase locally the availability of free electrons and holes in solid electrolytes. Factoring effects arising from extended defects appear crucial to explain electrochemical and $-$mechanical observations in ASSBs.

cond-mat.mtrl-sci↗

arXiv2023

Demystifying Digital Twin Buzzword: A Novel Generic Evaluation Model

Despite the growing popularity of digital twin (DT) developments, there is a lack of common understanding and definition for important concepts of DT. It is needed to address this gap by building a shared understanding of DT before it becomes an obstacle for future work. With this challenge in view, the objective of our study is to assess the existing DT from various domains on a common basis and to unify the knowledge and understanding of DT developers and stakeholders before practice. To achieve this goal, we conducted a systematic literature review and analyzed 25 selected papers to identify and discuss the characteristics of existing DT's. The review shows an inconsistency and case-specific choices of dimensions in assessing DT. Therefore, this article proposes a four-dimensional evaluation framework to assess the maturity of digital twins across different domains, focusing on the characteristics of digital models. The four identified dimensions in this model are Capability, Cooperability, Coverage, and Lifecycle. Additionally, a weight mechanism is implemented inside the model to adapt the importance of each dimension for different application requirements. Several case studies are devised to validate the proposed model in general, industrial and scientific cases.

cs.SE↗

arXiv2022

Physics-driven Deep Learning Inversion for Direct Current Resistivity Survey Data

The direct-current (DC) resistivity method is a commonly used geophysical technique for surveying adverse geological conditions. Inversion can reconstruct the resistivity model from data, which is an important step in the geophysical survey. However, the inverse problem is a serious ill-posed problem that makes it easy to obtain incorrect inversion results. Deep learning (DL) provides new avenues for solving inverse problems, and has been widely studied. Currently, most DL inversion methods for resistivity are purely data-driven and depend heavily on labels (real resistivity models). However, real resistivity models are difficult to obtain through field surveys. An inversion network may not be effectively trained without labels. In this study, we built an unsupervised learning resistivity inversion scheme based on the physical law of electric field propagation. First, a forward modeling process was embedded into the network training, which converted the predicted model to predicted data and formed a data misfit to the observation data. Unsupervised training independent of the real model was realized using the data misfit as a loss function. Moreover, a dynamic smoothing constraint was imposed on the loss function to alleviate the ill-posed inverse problem. Finally, a transfer learning scheme was applied to adapt the trained network with simulated data to field data. Numerical simulations and field tests showed that the proposed method can accurately locate and depict geological targets.

physics.geo-ph↗

arXiv2022

Adaptive dynamic programming for nonaffine nonlinear optimal control problem with state constraints

This paper presents a constrained adaptive dynamic programming (CADP) algorithm to solve general nonlinear nonaffine optimal control problems with known dynamics. Unlike previous ADP algorithms, it can directly deal with problems with state constraints. Firstly, a constrained generalized policy iteration (CGPI) framework is developed to handle state constraints by transforming the traditional policy improvement process into a constrained policy optimization problem. Next, we propose an actor-critic variant of CGPI, called CADP, in which both policy and value functions are approximated by multi-layer neural networks to directly map the system states to control inputs and value function, respectively. CADP linearizes the constrained optimization problem locally into a quadratically constrained linear programming problem, and then obtains the optimal update of the policy network by solving its dual problem. A trust region constraint is added to prevent excessive policy update, thus ensuring linearization accuracy. We determine the feasibility of the policy optimization problem by calculating the minimum trust region boundary and update the policy using two recovery rules when infeasible. The vehicle control problem in the path-tracking task is used to demonstrate the effectiveness of this proposed method.

eess.SY↗

arXiv2022

Recurrent Model Predictive Control: Learning an Explicit Recurrent Controller for Nonlinear Systems

This paper proposes an offline control algorithm, called Recurrent Model Predictive Control (RMPC), to solve large-scale nonlinear finite-horizon optimal control problems. It can be regarded as an explicit solver of traditional Model Predictive Control (MPC) algorithms, which can adaptively select appropriate model prediction horizon according to current computing resources, so as to improve the policy performance. Our algorithm employs a recurrent function to approximate the optimal policy, which maps the system states and reference values directly to the control inputs. The output of the learned policy network after N recurrent cycles corresponds to the nearly optimal solution of N-step MPC. A policy optimization objective is designed by decomposing the MPC cost function according to the Bellman's principle of optimality. The optimal recurrent policy can be obtained by directly minimizing the designed objective function, which is applicable for general nonlinear and non input-affine systems. Both simulation-based and real-robot path-tracking tasks are utilized to demonstrate the effectiveness of the proposed method.

eess.SY↗

arXiv2022

Topology optimization on complex surfaces based on the moving morphable component (MMC) method and computational conformal mapping (CCM)

In the present paper, an integrated paradigm for topology optimization on complex surfaces with arbitrary genus is proposed. The approach is constructed based on the two-dimensional (2D) Moving Morphable Component (MMC) framework, where a set of structural components are used as the basic units of optimization, and computational conformal mapping (CCM) technique, with which a complex surface represented by an unstructured triangular mesh can be mapped into a set of regular 2D parameter domains numerically. A multi-patch stitching scheme is also developed to achieve an MMC-friendly global parameterization through a number of local parameterizations. Numerical examples including a saddle-shaped shell, a torus-shape shell and a tee-branch pipe are solved to demonstrate the validity and efficiency of the proposed approach. It is found that compared with traditional approaches for topology optimization on 2D surfaces, optimized designs with clear load transmission paths can be obtained with much fewer numbers of design variables and degrees of freedom for finite element analysis (FEA) via the proposed approach.

math.OC↗

arXiv2021

Recurrent Model Predictive Control

This paper proposes an off-line algorithm, called Recurrent Model Predictive Control (RMPC), to solve general nonlinear finite-horizon optimal control problems. Unlike traditional Model Predictive Control (MPC) algorithms, it can make full use of the current computing resources and adaptively select the longest model prediction horizon. Our algorithm employs a recurrent function to approximate the optimal policy, which maps the system states and reference values directly to the control inputs. The number of prediction steps is equal to the number of recurrent cycles of the learned policy function. With an arbitrary initial policy function, the proposed RMPC algorithm can converge to the optimal policy by directly minimizing the designed loss function. We further prove the convergence and optimality of the RMPC algorithm thorough Bellman optimality principle, and demonstrate its generality and efficiency using two numerical examples.

eess.SY↗

arXiv2019

Stability Analysis of Interface Conditions for Ocean-Atmosphere Coupling

In this paper we analyze the stability of different coupling strategies for multidomain PDEs that arise in general circulation models used in climate simulations. We focus on fully coupled ocean-atmosphere models that are needed to represent and understand the complicated interactions of these two systems, becoming increasingly important in climate change assessment in recent years. Numerical stability issues typically arise because of different time-stepping strategies applied to the coupled PDE system. In particular, the contributing factors include using large time steps, lack of accurate interface flux, and singe-iteration coupling. We investigate the stability of the coupled ocean-atmosphere models for various interface conditions such as the Dirichlet-Neumann condition and the bulk interface condition, which is unique to climate modeling. By analyzing a simplified model, we demonstrate how the parameterization of the bulk condition and other numerical and physical parameters affect the coupling stability.

math.NA↗