arXiv subjects

Zhi Luo

Publications and source records attributed to Zhi Luo.

14 recordsLinked to original sources

arXiv2026

Artifact-Conditioned Interval Diagnostics for Flow-Matching Neural Posterior Estimation in a Controlled Gravitational-Wave Benchmark

Calibration checks for neural posterior estimators in gravitational-wave inference should remain interpretable when observations contain data-quality artifacts. We study marginal interval calibration in a controlled frequency-domain binary-black-hole benchmark with synthetic glitches, frequency masks, and power spectral density (PSD) mismatch. The posterior sampler is a support-aware flow-matching posterior estimator (FMPE) with a circular representation of coalescence phase. We compare raw marginal credible intervals with global rescaling, oracle artifact-stratified rescaling, hard predicted-label rescaling, and soft learned artifact-aware interval rescaling (LAIR). In the 1024-bin evaluation, a single global scale fitted on mixed calibration data transfers poorly to frequency-mask cases, giving a mean absolute 90\% marginal coverage error (MA90CE) of 0.1195. Soft LAIR lowers the corresponding error to 0.0672, although it is not uniformly better than the raw FMPE intervals. A 40-split LAIR evaluation and a six-checkpoint FMPE training-seed study show that the frequency-mask behavior is not a single-split artifact. The classifier recognizes frequency masks and PSD mismatch reliably, while glitch recall remains low. Waveform-resolution tests, PyCBC/LAL TaylorF2 backend checks, prior and Gaussian baselines, and controlled-likelihood reference-posterior probes indicate that marginal coverage must be read together with posterior width, geometry, and likelihood-based diagnostics. These results support using LAIR as an artifact-structured interval diagnostic, not as a substitute for posterior validation.

astro-ph.IM↗

arXiv2025

An Image Is Worth Ten Thousand Words: Verbose-Text Induction Attacks on VLMs

With the remarkable success of Vision-Language Models (VLMs) on multimodal tasks, concerns regarding their deployment efficiency have become increasingly prominent. In particular, the number of tokens consumed during the generation process has emerged as a key evaluation metric.Prior studies have shown that specific inputs can induce VLMs to generate lengthy outputs with low information density, which significantly increases energy consumption, latency, and token costs. However, existing methods simply delay the occurrence of the EOS token to implicitly prolong output, and fail to directly maximize the output token length as an explicit optimization objective, lacking stability and controllability.To address these limitations, this paper proposes a novel verbose-text induction attack (VTIA) to inject imperceptible adversarial perturbations into benign images via a two-stage framework, which identifies the most malicious prompt embeddings for optimizing and maximizing the output token of the perturbed images.Specifically, we first perform adversarial prompt search, employing reinforcement learning strategies to automatically identify adversarial prompts capable of inducing the LLM component within VLMs to produce verbose outputs. We then conduct vision-aligned perturbation optimization to craft adversarial examples on input images, maximizing the similarity between the perturbed image's visual embeddings and those of the adversarial prompt, thereby constructing malicious images that trigger verbose text generation. Comprehensive experiments on four popular VLMs demonstrate that our method achieves significant advantages in terms of effectiveness, efficiency, and generalization capability.

cs.CV↗

arXiv2025

PolicyEvolve: Evolving Programmatic Policies by LLMs for multi-player games via Population-Based Training

Multi-agent reinforcement learning (MARL) has achieved significant progress in solving complex multi-player games through self-play. However, training effective adversarial policies requires millions of experience samples and substantial computational resources. Moreover, these policies lack interpretability, hindering their practical deployment. Recently, researchers have successfully leveraged Large Language Models (LLMs) to generate programmatic policies for single-agent tasks, transforming neural network-based policies into interpretable rule-based code with high execution efficiency. Inspired by this, we propose PolicyEvolve, a general framework for generating programmatic policies in multi-player games. PolicyEvolve significantly reduces reliance on manually crafted policy code, achieving high-performance policies with minimal environmental interactions. The framework comprises four modules: Global Pool, Local Pool, Policy Planner, and Trajectory Critic. The Global Pool preserves elite policies accumulated during iterative training. The Local Pool stores temporary policies for the current iteration; only sufficiently high-performing policies from this pool are promoted to the Global Pool. The Policy Planner serves as the core policy generation module. It samples the top three policies from the Global Pool, generates an initial policy for the current iteration based on environmental information, and refines this policy using feedback from the Trajectory Critic. Refined policies are then deposited into the Local Pool. This iterative process continues until the policy achieves a sufficiently high average win rate against the Global Pool, at which point it is integrated into the Global Pool. The Trajectory Critic analyzes interaction data from the current policy, identifies vulnerabilities, and proposes directional improvements to guide the Policy Planner

cs.LG↗

arXiv2024

Provably Efficient Action-Manipulation Attack Against Continuous Reinforcement Learning

Manipulating the interaction trajectories between the intelligent agent and the environment can control the agent's training and behavior, exposing the potential vulnerabilities of reinforcement learning (RL). For example, in Cyber-Physical Systems (CPS) controlled by RL, the attacker can manipulate the actions of the adopted RL to other actions during the training phase, which will lead to bad consequences. Existing work has studied action-manipulation attacks in tabular settings, where the states and actions are discrete. As seen in many up-and-coming RL applications, such as autonomous driving, continuous action space is widely accepted, however, its action-manipulation attacks have not been thoroughly investigated yet. In this paper, we consider this crucial problem in both white-box and black-box scenarios. Specifically, utilizing the knowledge derived exclusively from trajectories, we propose a black-box attack algorithm named LCBT, which uses the Monte Carlo tree search method for efficient action searching and manipulation. Additionally, we demonstrate that for an agent whose dynamic regret is sub-linearly related to the total number of steps, LCBT can teach the agent to converge to target policies with only sublinear attack cost, i.e., $O\left(\mathcal{R}(T) + MH^3K^E\log (MT)\right)(0<E<1)$, where $H$ is the number of steps per episode, $K$ is the total number of episodes, $T=KH$ is the total number of steps, $M$ is the number of subspaces divided in the state space, and $\mathcal{R}(T)$ is the bound of the RL algorithm's regret. We conduct our proposed attack methods on three aggressive algorithms: DDPG, PPO, and TD3 in continuous settings, which show a promising attack performance.

cs.LG↗

arXiv2024

Holographic Einstein Rings of AdS Black Holes in Horndeski Theory

By utilizing the AdS/CFT correspondence and wave optics techniques, we conducted an extensive study of the imaging properties of holographic Einstein rings in the context of Anti-de Sitter (AdS) black holes (BHs) in Horndeski theory. Our results indicate that the optical characteristics of these holographic Einstein rings are significantly influenced by the observer's position, the physical parameters of the BH, the nature of the wave source, and the configuration of the optical system. Specifically, when the observer is positioned at the north pole of the AdS boundary, the holographic image prominently displays a ring structure aligning with the BH's photon sphere. We thoroughly analyzed how various physical parameters -- including the observation position, event horizon radius, temperature, and the parameter $\gamma$ in Horndeski theory -- affect the holographic Einstein rings. These parameters play a crucial role in determining the rings' radius and brightness, with variations potentially causing the ring structures to deform or even transform into bright spots. Furthermore, our comparative analysis between wave optics and geometric optics reveals a strong agreement in predicting the positions and brightnesses of both the photon ring and the Einstein ring. This research offers new insights into the spacetime geometry of BHs in Horndeski theory and proposes a promising framework for exploring the gravitational duals of strongly coupled systems.

gr-qc↗

arXiv2024

Shadows, Quasinormal Modes, and Optical Appearances of Black Holes in Horndeski Theory

This work describes the motion of photons in black hole (BH) spacetimes within the framework of Horndeski theory. We focus on the shadows, quasinormal modes (QNMs) and optical appearances of BHs surrounded by geometrically thin accretion disks. The QNMs of BHs are calculated by the WKB method and the eikonal limit, respectively. Using Event Horizon Telescope (EHT) observations of $\mathrm{M} 87^*$ and $\mathrm{Sgr} \mathrm{A}^*$, we can constrain the parameter in Horndeski theory to a small range. Based on the constraint, we obtain the frequency ranges of the fundamental modes for $\mathrm{M} 87^*$ and $\mathrm{Sgr} \mathrm{A}^*$ in Horndeski theory. By exploring the optical appearances of BHs, we find that for the current resolution of the EHT, it primarily captures direct emission. This work advances our understanding of the observational characteristics of BHs in Horndeski theory and constrains Horndeski theory by EHT observations of $\mathrm{M} 87^*$ and $\mathrm{Sgr} \mathrm{A}^*$.

gr-qc↗

arXiv2024

Amino Acids Stabilizing Effect on Protein and Colloidal Dispersions

Despite being used for decades as stabilizers, amino acids (AAs) remain mysterious components of many medical and biological formulations. Hypotheses on their role vary ranging from hydrotropic to protein-specific effects (stabilization against misfolding). Here, we deduce that AAs possess a new and broad colloidal property by finding that stabilizing effect of the AAs is comparable on dispersion of various proteins, plasmid DNA, and non-biological nanoparticles. The interactions among colloidal particles in dispersion are carefully evaluated by the second osmotic virial coefficient (B_22) and the potential of mean force. We propose a theoretical framework that explains the stabilization as the effect of weakly interacting small molecules with patchy nanoscale colloids. We validate it through quantitative comparison with experimental data by comparing equilibrium dissociation constants for AA/proteins obtained either by fitting the B22 data with this theory or experimentally. We find excellent quantitative agreement (e.g. proline/lysozyme 1.18 and 2.28 M, respectively) and indeed that the interactions are very weak. The theory presented implies that (i) charged AAs will be effective only for proteins of opposite charge; (ii) short peptides composed of n AAs will be as or more effective than n separate AAs; (iii) any small molecule weakly interacting with nanoscale colloids that increases the solvation of the surface will have a stabilizing effect. The experimental evidences corroborate all three predictions. Much like the ionic strength of the solution is commonly reported, our results imply that the same should be done for the small molecules, as they also affect fundamentally colloidal properties. As an example, we show that AAs vary the cloud point of a lysozyme solution by as much as 4 K.

physics.chem-ph↗

arXiv2023

Observational appearance and additional photon rings of the asymmetric thin-shell wormhole in Horndeski theory

In this paper, we study the observational appearance of the asymmetric thin-shell wormhole (ATW) in Horndeski theory by employing the ray-tracing method. We first calculate the effective potential and null geodesic of the ATW, and then we obtain the deflection angle of the photon in the ATW spacetime. Based on the impact parameter of the photon, the trajectory of the photon can be classified into three cases. Two typical emission models of the thin accretion disk are considered to analyze the observational appearance of the ATW. By comparing the observational appearances of the ATW and a black hole with the same mass parameter, we find additional features in the observational appearance of the ATW, such as the ``lensing band" and ``photon ring group".

gr-qc↗

arXiv2023

Shadow thermodynamics of the Hayward-AdS black hole

In this paper, the phase structure of the Hayward-AdS black hole (BH) is studied using shadow formalism. It has been found that the shadow radius is a monotonic function of the horizon radius and can therefore play an equivalent role to the horizon radius in characterizing the thermodynamics of Hayward-AdS BH. The thermodynamic phase transition (PT) of the Hayward-AdS BH is investigated with the shadow radius. It is shown that as the magnetic charge increases, the shadow radius becomes larger, while the coexistence temperature becomes lower. The thermal profile of the Hayward-AdS BH is established by combining the temperature diagram and the shadow cast diagram, which shows that for a fixed magnetic charge, the temperature of the Hayward-AdS BH increases with the pressure while the region of the thermal profile decreases with the pressure. In particular, the temperature of the Hayward-AdS BH follows an N-type change trend when it is smaller than the critical temperature. This imply that the BH shadow may be used to investigate the thermodynamics of the Hayward-AdS BH.

gr-qc↗

arXiv2022

Effects of a global monopole on thermodynamic phase transition of the charged AdS black hole

In this paper, we study the dynamical properties of thermodynamic phase transition (PT) for the charged AdS black hole (BH) with a global monopole via the Gibbs free energy landscape and reveal the effects of a global monopole on the kinetics of the AdS BH thermodynamic PT. First, we briefly review the thermodynamics of the charged AdS BH with a global monopole. Then, we introduce the Gibbs free energy landscape to study the thermodynamic stability of the BH state. Because of thermal fluctuations, the small black hole (SBH) state can transit to the large black hole (LBH) state, and vice versa. We use the Fokker-Planck equation with the reflecting boundary condition to study the probability evolution of the BH state with and without a global monopole separately. We find that for both the SBH and LBH states, the global monopole could slow down the evolution of the BH state. In addition, we obtain the relationship between the first passage time and the monopole parameter $η$. The result shows that as the monopole parameter $η$ increases, the mean first passage time will be longer for both the SBH and LBH states.

gr-qc↗

arXiv2021

On the correspondence between energy conservation and energy-momentum tensor conservation in cosmology

The correspondence between the thermodynamic energy equation satisfied by a closed co-moving volume and the conservation equation satisfied by the energy-momentum tensor of the matter inside the co-moving volume is extended to a more general system with an arbitrary cosmological horizon and a heat source. The energy of the system consisting of a cosmological horizon and its internal matter could be conserved by defining a surface energy on the horizons. Therefore, energy conservation and energy-momentum tensor conservation can always be consistent for such a system. On the other hand, from the perspective of classical thermodynamics, one can define an effective pressure at the cosmological horizon to guarantee that the thermodynamic energy equation inside the horizon is consistent with the energy-momentum tensor conservation equation of the matter inside the horizon. These systems can satisfy the generalized second law of thermodynamics under appropriate conditions. The definitions of the surface energy and the effective pressure are extended to the gravity theory with non-minimal coupling between geometry and matter, in which geometry could be regarded as a heat source.

gr-qc↗

arXiv2020

u-net CNN based fourier ptychography

Fourier ptychography is a recently explored imaging method for overcoming the diffraction limit of conventional cameras with applications in microscopy and yielding high-resolution images. In order to splice together low-resolution images taken under different illumination angles of coherent light source, an iterative phase retrieval algorithm is adopted. However, the reconstruction procedure is slow and needs a good many of overlap in the Fourier domain for the continuous recorded low-resolution images and is also worse under system aberrations such as noise or random update sequence. In this paper, we propose a new retrieval algorithm that is based on convolutional neural networks. Once well trained, our model can perform high-quality reconstruction rapidly by using the graphics processing unit. The experiments demonstrate that our model achieves better reconstruction results and is more robust under system aberrations.

eess.IV↗

arXiv2020

Inverse design of multilayer nanoparticles using artificial neural networks and genetic algorithm

The light scattering of multilayer nanoparticles can be solved by Maxwell equations. However, it is difficult to solve the inverse design of multilayer nanoparticles by using the traditional trial-and-error method. Here, we present a method for forward simulation and inverse design of multilayer nanoparticles. We combine the global search ability of genetic algorithm with the local search ability of neural network. First, the genetic algorithm is used to find a suitable solution, and then the neural network is used to fine-tune it. Due to the non-unique relationship between physical structures and optical responses, we first train a forward neural network, and then it is applied to the inverse design of multilayer nanoparticles. Not only here, this method can easily be extended to predict and find the best design parameters for other optical structures.

eess.SP↗

arXiv2015

Multi-level Resistive Switching Characteristics of W/Co:TiO2/FTO Structures

In the present work, multi-level resistive switching (RS) in W/Co:TiO2/FTO structures induced by a multi-mixed mechanism was studied. It was found that the devices could be reproducibly programmed into three nonvolatile resistance states. And the directly switching between any resistance states was realized. This increases the operation speed and lowers the complexity of control circuit of multi-state nonvolatile memory.

cond-mat.mtrl-sci↗