arXiv subjects

Peng Liu

Publications and source records attributed to Peng Liu.

At least 307 records · Page 17Linked to original sources

arXiv2022

RSSD: Defend against Ransomware with Hardware-Isolated Network-Storage Codesign and Post-Attack Analysis

Encryption ransomware has become a notorious malware. It encrypts user data on storage devices like solid-state drives (SSDs) and demands a ransom to restore data for users. To bypass existing defenses, ransomware would keep evolving and performing new attack models. For instance, we identify and validate three new attacks, including (1) garbage-collection (GC) attack that exploits storage capacity and keeps writing data to trigger GC and force SSDs to release the retained data; (2) timing attack that intentionally slows down the pace of encrypting data and hides its I/O patterns to escape existing defense; (3) trimming attack that utilizes the trim command available in SSDs to physically erase data. To enhance the robustness of SSDs against these attacks, we propose RSSD, a ransomware-aware SSD. It redesigns the flash management of SSDs for enabling the hardware-assisted logging, which can conservatively retain older versions of user data and received storage operations in time order with low overhead. It also employs hardware-isolated NVMe over Ethernet to expand local storage capacity by transparently offloading the logs to remote cloud/servers in a secure manner. RSSD enables post-attack analysis by building a trusted evidence chain of storage operations to assist the investigation of ransomware attacks. We develop RSSD with a real-world SSD FPGA board. Our evaluation shows that RSSD can defend against new and future ransomware attacks, while introducing negligible performance overhead.

cs.CR↗

arXiv2022

Dynamical spontaneous scalarization in Einstein-Maxwell-scalar models in anti-de Sitter spacetime

The phenomenon of spontaneous scalarization of charged black holes has attracted a lot of attention. In this work, we study the dynamical process of the spontaneous scalarization of charged black hole in asymptotically anti-de Sitter spacetimes in Einstein-Maxwell-scalar models. Including various non-minimal couplings between the scalar field and Maxwell field, we observe that an initial scalar-free configuration suffers tachyonic instability and both the scalar field and the black hole irreducible mass grow exponentially at early times and saturate exponentially at late times. For fractional couplings, we find that though there is negative energy distribution near the black hole horizon, the black hole horizon area never decreases. But when the parameters are large, the evolution endpoints of linearly unstable bald black holes will be spacetimes with naked singularity and the cosmic censorship is violated. The effects of the black hole charge, cosmological constant and coupling strength on the dynamical scalarization process are studied in detail. We find that large enough cosmological constant can prevent the spontaneous scalarization.

gr-qc↗

arXiv2022

Towards Understanding and Harnessing the Effect of Image Transformation in Adversarial Detection

Deep neural networks (DNNs) are threatened by adversarial examples. Adversarial detection, which distinguishes adversarial images from benign images, is fundamental for robust DNN-based services. Image transformation is one of the most effective approaches to detect adversarial examples. During the last few years, a variety of image transformations have been studied and discussed to design reliable adversarial detectors. In this paper, we systematically synthesize the recent progress on adversarial detection via image transformations with a novel classification method. Then, we conduct extensive experiments to test the detection performance of image transformations against state-of-the-art adversarial attacks. Furthermore, we reveal that each individual transformation is not capable of detecting adversarial examples in a robust way, and propose a DNN-based approach referred to as \emph{AdvJudge}, which combines scores of 9 image transformations. Without knowing which individual scores are misleading or not misleading, AdvJudge can make the right judgment, and achieve a significant improvement in detection rate. Finally, we utilize an explainable AI tool to show the contribution of each image transformation to adversarial detection. Experimental results show that the contribution of image transformations to adversarial detection is significantly different, the combination of them can significantly improve the generic detection ability against state-of-the-art adversarial attacks.

cs.CV↗

arXiv2022

Nebula-I: A General Framework for Collaboratively Training Deep Learning Models on Low-Bandwidth Cloud Clusters

The ever-growing model size and scale of compute have attracted increasing interests in training deep learning models over multiple nodes. However, when it comes to training on cloud clusters, especially across remote clusters, huge challenges are faced. In this work, we introduce a general framework, Nebula-I, for collaboratively training deep learning models over remote heterogeneous clusters, the connections between which are low-bandwidth wide area networks (WANs). We took natural language processing (NLP) as an example to show how Nebula-I works in different training phases that include: a) pre-training a multilingual language model using two remote clusters; and b) fine-tuning a machine translation model using knowledge distilled from pre-trained models, which run through the most popular paradigm of recent deep learning. To balance the accuracy and communication efficiency, in Nebula-I, parameter-efficient training strategies, hybrid parallel computing methods and adaptive communication acceleration techniques are jointly applied. Meanwhile, security strategies are employed to guarantee the safety, reliability and privacy in intra-cluster computation and inter-cluster communication. Nebula-I is implemented with the PaddlePaddle deep learning framework, which can support collaborative training over heterogeneous hardware, e.g. GPU and NPU. Experiments demonstrate that the proposed framework could substantially maximize the training efficiency while preserving satisfactory NLP performance. By using Nebula-I, users can run large-scale training tasks over cloud clusters with minimum developments, and the utility of existed large pre-trained models could be further promoted. We also introduced new state-of-the-art results on cross-lingual natural language inference tasks, which are generated based upon a novel learning framework and Nebula-I.

cs.LG↗

arXiv2022

Dynamical spontaneous scalarization in Einstein-Maxwell-scalar theory

We study the linear instability and the nonlinear dynamical evolution of the Reissner-Nordström (RN) black hole in the Einstein-Maxwell-scalar theory in asymptotic flat spacetime. We focus on the coupling function $f(ϕ)=e^{-bϕ^2}$ which allows both the scalar-free RN solution and scalarized black hole solution. We first present the evolution of system parameters during dynamic scalarization. For parameter regions where spontaneous scalarization occurs, we find that the evolution of the scalar field at the horizon is dominated by the fundamental unstable mode from linear analysis at early times. At late times, the nonlinear evolution can be viewed as the perturbation of scalarized black holes.

gr-qc↗

arXiv2022

Balancing Multi-Domain Corpora Learning for Open-Domain Response Generation

Open-domain conversational systems are assumed to generate equally good responses on multiple domains. Previous work achieved good performance on the single corpus, but training and evaluating on multiple corpora from different domains are less studied. This paper explores methods of generating relevant responses for each of multiple multi-domain corpora. We first examine interleaved learning which intermingles multiple corpora as the baseline. We then investigate two multi-domain learning methods, labeled learning and multi-task labeled learning, which encode each corpus through a unique corpus embedding. Furthermore, we propose Domain-specific Frequency (DF), a novel word-level importance weight that measures the relative importance of a word for a specific corpus compared to other corpora. Based on DF, we propose weighted learning, a method that integrates DF to the loss function. We also adopt DF as a new evaluation metric. Extensive experiments show that our methods gain significant improvements on both automatic and human evaluation. We share our code and data for reproducibility

cs.CL↗

arXiv2022

A novel holographic quantum phase transition and butterfly velocity

In this paper, we make a systematical and in-depth exploration on the phase structure and the behaviors of butterfly velocity in an Einstein-Maxwell-dilaton-axions (EMDA) model. Depending on the model parameter, there are two kinds of mechanisms driving quantum phase transition (QPT) in this model. One is the infrared (IR) geometry to be renormalization group (RG) unstable, and the other is the strength of lattice deformation leading to some kind of bifurcating solution. We also find a novel QPT in the metal phases. The study on the behavior of the butterfly velocity crossing QPT indicates that the butterfly velocity or its first derivative exhibiting local extreme depends on the QPT mechanism. Further, the scaling behaviors of the butterfly velocity in the zero-temperature limit confirm that different phases are controlled by different IR geometries. Therefore, the butterfly velocity is a good probe to QPT and it also provides a possible way to study QPT beyond holography.

hep-th↗

arXiv2022

Semantic-preserving Reinforcement Learning Attack Against Graph Neural Networks for Malware Detection

As an increasing number of deep-learning-based malware scanners have been proposed, the existing evasion techniques, including code obfuscation and polymorphic malware, are found to be less effective. In this work, we propose a reinforcement learning-based semantics-preserving (i.e.functionality-preserving) attack against black-box GNNs (GraphNeural Networks) for malware detection. The key factor of adversarial malware generation via semantic Nops insertion is to select the appropriate semanticNopsand their corresponding basic blocks. The proposed attack uses reinforcement learning to automatically make these "how to select" decisions. To evaluate the attack, we have trained two kinds of GNNs with five types(i.e., Backdoor, Trojan-Downloader, Trojan-Ransom, Adware, and Worm) of Windows malware samples and various benign Windows programs. The evaluation results have shown that the proposed attack can achieve a significantly higher evasion rate than three baseline attacks, namely the semantics-preserving random instruction insertion attack, the semantics-preserving accumulative instruction insertion attack, and the semantics-preserving gradient-based instruction insertion attack.

cs.CR↗

arXiv2022

Optical Wavelength Meter with Machine Learning Enhanced Precision

Diverse applications in photonics and microwave engineering require a means of measurement of the instantaneous frequency of a signal. A photonic implementation typically applies an interferometer equipped with three or more output ports to measure the frequency dependent phase shift provided by an optical delay line. The components constituting the interferometer are prone to impairments which results in erroneous measurements. It is shown that the information to be retrieved is encoded by a three-component vector that lies on a circular cone within a three-dimensional Cartesian object space. The measured data belongs to the image of the object space under a linear map that describes the action of the interferometer. Assisted by a learning algorithm, an inverse map from the image space into the object space is constructed. The inverse map compensates for a variety of impairments while being robust to noise. Simulation results demonstrate that, to the extent the interferometer model captures all significant impairments, a precision limited only by the level of random noise is attainable. A wavelength meter architecture is fabricated on Si3N4 photonic integration platform to prove the method experimentally. Applied to the measured data, greater than an order of magnitude improvement in precision is achieved by the proposed method compared to the conventional method.

physics.optics↗

arXiv2022

Mixed State Entanglement For Holographic Systems With A Scalar Hair

We study the mixed state entanglement of an asymptotic AdS black hole system with scalar hair. Through numerical calculations, we find that the holographic entanglement entropy (HEE) presents a non-monotonic behavior with the system parameter, depending on the size of the subregion. In addition, the mutual information (MI) also shows non-monotonic behavior in certain ranges of system parameters. However, the entanglement wedge minimum cross-section (EWCS), which is a mixed state entanglement measure, increases monotonically with the AdS radius; meanwhile, shows non-monotonic with the temperature. We also give analytical understandings of the phenomena above.

hep-th↗

arXiv2022

Pessimistic Bootstrapping for Uncertainty-Driven Offline Reinforcement Learning

Offline Reinforcement Learning (RL) aims to learn policies from previously collected datasets without exploring the environment. Directly applying off-policy algorithms to offline RL usually fails due to the extrapolation error caused by the out-of-distribution (OOD) actions. Previous methods tackle such problem by penalizing the Q-values of OOD actions or constraining the trained policy to be close to the behavior policy. Nevertheless, such methods typically prevent the generalization of value functions beyond the offline data and also lack precise characterization of OOD data. In this paper, we propose Pessimistic Bootstrapping for offline RL (PBRL), a purely uncertainty-driven offline algorithm without explicit policy constraints. Specifically, PBRL conducts uncertainty quantification via the disagreement of bootstrapped Q-functions, and performs pessimistic updates by penalizing the value function based on the estimated uncertainty. To tackle the extrapolating error, we further propose a novel OOD sampling method. We show that such OOD sampling and pessimistic bootstrapping yields provable uncertainty quantifier in linear MDPs, thus providing the theoretical underpinning for PBRL. Extensive experiments on D4RL benchmark show that PBRL has better performance compared to the state-of-the-art algorithms.

cs.LG↗

arXiv2022

Dynamic Properties of Two-Dimensional Latticed Holographic System

We study the anisotropic properties of dynamical quantities: direct current (DC) conductivity, butterfly velocity, and charge diffusion. The anisotropy plays a crucial role in determining the phase structure of the two-lattice system. Even a small deviation from isotropy can lead to distinct phase structures, as well as the IR fixed points of our holographic systems. In particular, for anisotropic cases, the most important property is that the IR fixed point can be non-AdS$_2 \times \mathbb R^2$ even for metallic phases. As that of a one-lattice system, the butterfly velocity and the charge diffusion can also diagnose the quantum phase transition (QPT) in this two-dimensional anisotropic latticed system.

hep-th↗

arXiv2022

Quasi-normal modes of the Einstein-Maxwell-aether Black Hole

We study the quasi-normal modes of the charged scalar perturbations in the background of the Einstein-Maxwell-aether black hole through three methods (WKB method, continued fraction method, generalized eigenvalue method). Then we propose the specific treatment for the generalized effective potential with $ω$-dependence and the complete procedure of transforming calculation continued fraction method into finding the zero point of the corresponding complex function numerically. These methods are valid because the results from different methods are consistent. We also investigate the allowed region of the second kind aether black hole among the system parameters ($c_{13},c_{14},Q$). Finally we show the existence of quasi-resonances of massive perturbation for Einstein-Maxwell-aether black hole even with large aether parameter.

gr-qc↗

arXiv2022

Single Model Deep Learning on Imbalanced Small Datasets for Skin Lesion Classification

Deep convolutional neural network (DCNN) models have been widely explored for skin disease diagnosis and some of them have achieved the diagnostic outcomes comparable or even superior to those of dermatologists. However, broad implementation of DCNN in skin disease detection is hindered by small size and data imbalance of the publically accessible skin lesion datasets. This paper proposes a novel single-model based strategy for classification of skin lesions on small and imbalanced datasets. First, various DCNNs are trained on different small and imbalanced datasets to verify that the models with moderate complexity outperform the larger models. Second, regularization DropOut and DropBlock are added to reduce overfitting and a Modified RandAugment augmentation strategy is proposed to deal with the defects of sample underrepresentation in the small dataset. Finally, a novel Multi-Weighted New Loss (MWNL) function and an end-to-end cumulative learning strategy (CLS) are introduced to overcome the challenge of uneven sample size and classification difficulty and to reduce the impact of abnormal samples on training. By combining Modified RandAugment, MWNL and CLS, our single DCNN model method achieved the classification accuracy comparable or superior to those of multiple ensembling models on different dermoscopic image datasets. Our study shows that this method is able to achieve a high classification performance at a low cost of computational resources and inference time, potentially suitable to implement in mobile devices for automated screening of skin lesions and many other malignancies in low resource settings.

cs.CV↗

arXiv2022

A Survey on Active Deep Learning: From Model-driven to Data-driven

Which samples should be labelled in a large data set is one of the most important problems for trainingof deep learning. So far, a variety of active sample selection strategies related to deep learning havebeen proposed in many literatures. We defined them as Active Deep Learning (ADL) only if theirpredictor is deep model, where the basic learner is called as predictor and the labeling schemes iscalled selector. In this survey, three fundamental factors in selector designation were summarized. Wecategory ADL into model-driven ADL and data-driven ADL, by whether its selector is model-drivenor data-driven. The different characteristics of the two major type of ADL were addressed in indetail respectively. Furthermore, different sub-classes of data-driven and model-driven ADL are alsosummarized and discussed emphatically. The advantages and disadvantages between data-driven ADLand model-driven ADL are thoroughly analyzed. We pointed out that, with the development of deeplearning, the selector in ADL also is experiencing the stage from model-driven to data-driven. Finally,we make discussion on ADL about its uncertainty, explanatory, foundations of cognitive science etc.and survey on the trend of ADL from model-driven to data-driven.

cs.LG↗

arXiv2022

Entanglement Wedge Minimum Cross-Section in Holographic Axion Gravity Theories

We study the mixed state entanglement properties in two holographic axion models by examining the behavior of the entanglement wedge minimum cross section (EWCS), and comparing it with the holographic entanglement entropy (HEE) and mutual information (MI). We find that the behavior of HEE, MI and EWCS with Hawking temperature is monotonic, while the behavior with the axion parameter $k$ is more rich, which depends on the size of the configuration and the values of the other two parameters. Interestingly, the EWCS monotonically increases with the coupling constant $κ$ between the axion field and the Maxwell field, while HEE and MI can be non-monotonic. It suggests that the EWCS, as a mixed state entanglement measure, captures distinct degrees of freedom from the HEE and MI indeed. We also provide analytical understandings for most of the numerical results.

hep-th↗

arXiv2022

Entanglement Wedge Minimum Cross-Section for Holographic Aether Gravity

We study the entanglement wedge cross-section (EWCS) in holographic Aether gravity theory, a gravity theory with Lorentz symmetry violation while keeping the general covariance intact. We find that only a limited parameter space is allowed to obtain a black brane with positive Hawking temperature. Subject to these allowed parameter regions, we find that the EWCS could exhibit non-monotonic behaviors with system parameters. Meanwhile, the holographic entanglement entropy (HEE), and the corresponding mutual information (MI), can only exhibit monotonic behaviors. These phenomena suggest that the EWCS could capture much more rich content of the entanglement than that of the HEE and the MI. The role of the Lorentz violation in determining the behaviors of quantum information-related quantities is also analyzed.

hep-th↗

arXiv2022

Reflected Entropy in Double Holography

Recently, the reflected entropy is proposed in holographic approach to describe the entanglement of a bipartite quantum system in a mixed state, which is identified as the area of the reflected minimal surface inside the entanglement wedge. In this paper, we study the reflected entropy in the doubly holographic setup, which contains the degrees of freedom of quantum matter in the bulk. In this context, we propose a notion of quantum entanglement wedge cross-section, which may describe the reflected entropy with higher-order quantum corrections. We numerically compute the reflected entropy in pure AdS background and black hole background in four dimensions, respectively. In general, the reflected entropy contains the contribution from the geometry on the brane and the contribution from the CFT. We compute their proportion for different Newton constants and find that their behaviors are in agreement with the results based on the semi-classical gravity and the correlation of CFT coupled to the bath CFT.

hep-th↗