arXiv subjects

Lei Pan

Publications and source records attributed to Lei Pan.

At least 73 records · Page 4Linked to original sources

arXiv2022

Towards Privacy-Preserving Neural Architecture Search

Machine learning promotes the continuous development of signal processing in various fields, including network traffic monitoring, EEG classification, face identification, and many more. However, massive user data collected for training deep learning models raises privacy concerns and increases the difficulty of manually adjusting the network structure. To address these issues, we propose a privacy-preserving neural architecture search (PP-NAS) framework based on secure multi-party computation to protect users' data and the model's parameters/hyper-parameters. PP-NAS outsources the NAS task to two non-colluding cloud servers for making full advantage of mixed protocols design. Complement to the existing PP machine learning frameworks, we redesign the secure ReLU and Max-pooling garbled circuits for significantly better efficiency ($3 \sim 436$ times speed-up). We develop a new alternative to approximate the Softmax function over secret shares, which bypasses the limitation of approximating exponential operations in Softmax while improving accuracy. Extensive analyses and experiments demonstrate PP-NAS's superiority in security, efficiency, and accuracy.

cs.CR↗

arXiv2022

Adaptive Cross-Attention-Driven Spatial-Spectral Graph Convolutional Network for Hyperspectral Image Classification

Recently, graph convolutional networks (GCNs) have been developed to explore spatial relationship between pixels, achieving better classification performance of hyperspectral images (HSIs). However, these methods fail to sufficiently leverage the relationship between spectral bands in HSI data. As such, we propose an adaptive cross-attention-driven spatial-spectral graph convolutional network (ACSS-GCN), which is composed of a spatial GCN (Sa-GCN) subnetwork, a spectral GCN (Se-GCN) subnetwork, and a graph cross-attention fusion module (GCAFM). Specifically, Sa-GCN and Se-GCN are proposed to extract the spatial and spectral features by modeling correlations between spatial pixels and between spectral bands, respectively. Then, by integrating attention mechanism into information aggregation of graph, the GCAFM, including three parts, i.e., spatial graph attention block, spectral graph attention block, and fusion block, is designed to fuse the spatial and spectral features and suppress noise interference in Sa-GCN and Se-GCN. Moreover, the idea of the adaptive graph is introduced to explore an optimal graph through back propagation during the training process. Experiments on two HSI data sets show that the proposed method achieves better performance than other classification methods.

cs.CV↗

arXiv2022

Spatial-Spectral Feature Extraction via Deep ConvLSTM Neural Networks for Hyperspectral Image Classification

In recent years, deep learning has presented a great advance in hyperspectral image (HSI) classification. Particularly, long short-term memory (LSTM), as a special deep learning structure, has shown great ability in modeling long-term dependencies in the time dimension of video or the spectral dimension of HSIs. However, the loss of spatial information makes it quite difficult to obtain the better performance. In order to address this problem, two novel deep models are proposed to extract more discriminative spatial-spectral features by exploiting the Convolutional LSTM (ConvLSTM). By taking the data patch in a local sliding window as the input of each memory cell band by band, the 2-D extended architecture of LSTM is considered for building the spatial-spectral ConvLSTM 2-D Neural Network (SSCL2DNN) to model long-range dependencies in the spectral domain. To better preserve the intrinsic structure information of the hyperspectral data, the spatial-spectral ConvLSTM 3-D Neural Network (SSCL3DNN) is proposed by extending LSTM to 3-D version for further improving the classification performance. The experiments, conducted on three commonly used HSI data sets, demonstrate that the proposed deep models have certain competitive advantages and can provide better classification performance than other state-of-the-art approaches.

cs.CV↗

arXiv2022

Video is All You Need: Attacking PPG-based Biometric Authentication

Unobservable physiological signals enhance biometric authentication systems. Photoplethysmography (PPG) signals are convenient owning to its ease of measurement and are usually well protected against remote adversaries in authentication. Any leaked PPG signals help adversaries compromise the biometric authentication systems, and the advent of remote PPG (rPPG) enables adversaries to acquire PPG signals through restoration. While potentially dangerous, rPPG-based attacks are overlooked because existing methods require the victim's PPG signals. This paper proposes a novel spoofing attack approach that uses the waveforms of rPPG signals extracted from video clips to fool the PPG-based biometric authentication. We develop a new PPG restoration model that does not require leaked PPG signals for adversarial attacks. Test results on state-of-art PPG-based biometric authentication show that the signals recovered through rPPG pose a severe threat to PPG-based biometric authentication.

cs.CR↗

arXiv2022

FAAG: Fast Adversarial Audio Generation through Interactive Attack Optimisation

Automatic Speech Recognition services (ASRs) inherit deep neural networks' vulnerabilities like crafted adversarial examples. Existing methods often suffer from low efficiency because the target phases are added to the entire audio sample, resulting in high demand for computational resources. This paper proposes a novel scheme named FAAG as an iterative optimization-based method to generate targeted adversarial examples quickly. By injecting the noise over the beginning part of the audio, FAAG generates adversarial audio in high quality with a high success rate timely. Specifically, we use audio's logits output to map each character in the transcription to an approximate position of the audio's frame. Thus, an adversarial example can be generated by FAAG in approximately two minutes using CPUs only and around ten seconds with one GPU while maintaining an average success rate over 85%. Specifically, the FAAG method can speed up around 60% compared with the baseline method during the adversarial example generation process. Furthermore, we found that appending benign audio to any suspicious examples can effectively defend against the targeted adversarial attack. We hope that this work paves the way for inventing new adversarial attacks against speech recognition with computational constraints.

cs.SD↗

arXiv2022

Composite Spin Approach to the Blockade Effect in Rydberg Atom Arrays

The Rydberg blockade induces strongly correlated many-body effects in Rydberg atom arrays, including rich ground-state phases and many-body scar states in the excitation spectrum. In this letter, we propose a composite spin representation that can provide a unified description for major features in this system. The composite spin combines Rydberg excitation and the auxiliary fermions, which are introduced to implement the Rydberg blockade constraint automatically. First, we focus on the PXP model describing one-dimensional arrays with the Rydberg blocking radius being a lattice spacing. Using composite spins, the ground state is simply a ferromagnetic product state of composite spins, and the magnon excitations of these composite spins can accurately describe the many-body scar states and signal the quantum phase transition driven by detuning. Then, we show that Rydberg atom arrays with different blocking radii can share a universal description of the composite spin representation, and the difference between different blockade radii can be absorbed in the formation of composite spins.

cond-mat.quant-gas↗

arXiv2021

NavP: Enabling Navigational Programming for Science Data Processing via Application-Initiated Checkpointing

Science Data Systems (SDS) handle science data from acquisition through processing to distribution. They are deployed in the Cloud today, and the efficiency of Cloud instance utilization is critical to success. Conventional SDS are unable to take advantage of a cost-effective Amazon EC2 spot market, especially for long-running tasks. Some of the difficulties found in current practice at NASA/JPL are: a lack of mechanism for app programmers to save valuable partial results for future processing continuation, the heavy weight from using container-based (Singularity) sandboxes with more than 200,000 OS-level files; and the gap between scientists developing algorithms/programs on a laptop and the SDS experts deploying software in Cloud computing or supercomputing. We present a first proof-of-principle of this using NavP (Navigational Programming) and fault-tolerant computing (FTC) in SDS, by employing program state migration facilitated by Checkpoint-Restart (C/R). NavP provides a new navigational view of computations in a distributed world for the application programmers. The tool of DHP (DMTCP Hop and Publish) we developed enables the application programmers to navigate the computation among instances or nodes by inserting hop(destination) statements in their app code, and choose when to publish partial results at stages of their algorithms that they think worthwhile for future continuation. The result of using DHP is that a parallel distributed SDS becomes easier to program and deploy, and this enables more efficient leveraging of the Amazon EC2 Spot market. This technical report describes a high-level design and an initial implementation.

cs.DC↗

arXiv2021

Machine Learning Based Cyber Attacks Targeting on Controlled Information: A Survey

Stealing attack against controlled information, along with the increasing number of information leakage incidents, has become an emerging cyber security threat in recent years. Due to the booming development and deployment of advanced analytics solutions, novel stealing attacks utilize machine learning (ML) algorithms to achieve high success rate and cause a lot of damage. Detecting and defending against such attacks is challenging and urgent so that governments, organizations, and individuals should attach great importance to the ML-based stealing attacks. This survey presents the recent advances in this new type of attack and corresponding countermeasures. The ML-based stealing attack is reviewed in perspectives of three categories of targeted controlled information, including controlled user activities, controlled ML model-related information, and controlled authentication information. Recent publications are summarized to generalize an overarching attack methodology and to derive the limitations and future directions of ML-based stealing attacks. Furthermore, countermeasures are proposed towards developing effective protections from three aspects -- detection, disruption, and isolation.

cs.CR↗

arXiv2021

ECG-Adv-GAN: Detecting ECG Adversarial Examples with Conditional Generative Adversarial Networks

Electrocardiogram (ECG) acquisition requires an automated system and analysis pipeline for understanding specific rhythm irregularities. Deep neural networks have become a popular technique for tracing ECG signals, outperforming human experts. Despite this, convolutional neural networks are susceptible to adversarial examples that can misclassify ECG signals and decrease the model's precision. Moreover, they do not generalize well on the out-of-distribution dataset. The GAN architecture has been employed in recent works to synthesize adversarial ECG signals to increase existing training data. However, they use a disjointed CNN-based classification architecture to detect arrhythmia. Till now, no versatile architecture has been proposed that can detect adversarial examples and classify arrhythmia simultaneously. To alleviate this, we propose a novel Conditional Generative Adversarial Network to simultaneously generate ECG signals for different categories and detect cardiac abnormalities. Moreover, the model is conditioned on class-specific ECG signals to synthesize realistic adversarial examples. Consequently, we compare our architecture and show how it outperforms other classification models in normal/abnormal ECG signal detection by benchmarking real world and adversarial signals.

cs.LG↗

arXiv2021

Quantum Many-Body Scars and Quantum Criticality

In this letter, we study the PXP Hamiltonian with an external magnetic field that exhibits both quantum scar states and quantum criticality. It is known that this model hosts a series of quantum many-body scar states violating quantum thermalization at zero magnetic field, and it also exhibits an Ising quantum phase transition driven by finite magnetic field. Although the former involves the properties of generic excited states and the latter concerns the low-energy physics, we discover two surprising connections between them, inspired by the observation that both states possess log-volume law entanglement entropies. First, we show that the quantum many-body scar states can be tracked to a set of quantum critical states, whose nature can be understood as pair-wisely occupied Fermi sea states. Second, we show that the partial violation of quantum thermalization diminishes in the quantum critical regime. We envision that these connections can be extended to general situations and readily verified in existing cold atom experimental platforms.

cond-mat.quant-gas↗

arXiv2021

Bulk dissipation in the quantum anomalous Hall effect

Even at the lowest accessible temperatures, measurements of the quantum anomalous Hall (QAH) effect have indicated the presence of parasitic dissipative conduction channels. There is no consensus whether parasitic conduction is related to processes in the bulk or along the edges. Here, we approach this problem by comparing transport measurements of Hall bar and Corbino geometry devices fabricated from Cr-doped (BiSb)$_2$Te$_3$. We identify bulk conduction as the dominant source of dissipation at all values of temperature and in-plane electric field. Furthermore, we observe identical breakdown phenomenology in both geometries, indicating that breakdown of the QAH phase is a bulk process. The methodology developed in this study could be used to identify dissipative conduction mechanisms in new QAH materials, ultimately guiding material development towards realization of the QAH effect at higher temperatures.

cond-mat.mes-hall↗

arXiv2021

The Audio Auditor: User-Level Membership Inference in Internet of Things Voice Services

With the rapid development of deep learning techniques, the popularity of voice services implemented on various Internet of Things (IoT) devices is ever increasing. In this paper, we examine user-level membership inference in the problem space of voice services, by designing an audio auditor to verify whether a specific user had unwillingly contributed audio used to train an automatic speech recognition (ASR) model under strict black-box access. With user representation of the input audio data and their corresponding translated text, our trained auditor is effective in user-level audit. We also observe that the auditor trained on specific data can be generalized well regardless of the ASR model architecture. We validate the auditor on ASR models trained with LSTM, RNNs, and GRU algorithms on two state-of-the-art pipelines, the hybrid ASR system and the end-to-end ASR system. Finally, we conduct a real-world trial of our auditor on iPhone Siri, achieving an overall accuracy exceeding 80\%. We hope the methodology developed in this paper and findings can inform privacy advocates to overhaul IoT privacy.

cs.CR↗

arXiv2021

Fate of symmetry protected coherence in open quantum system

We investigate the fate of coherence in the dynamical evolution of a symmetry protected quantum system. Under the formalism of system-plus-bath for open quantum system, the anti-unitary symmetry exhibits significant difference from the unitary one in protecting initial coherence. Specifically, taking advantage of Lindblad master equation, we find that a pure state in the symmetry protected degenerate subspace will decohere even though both the system Hamiltonian and system-environment interaction respect the same anti-unitary symmetry. In contrast, the coherence will persist when the protecting symmetry is unitary. We provide an elaborate classification table to illustrate what kinds of symmetry combinations are able to preserve the coherence of initial state, which is confirmed by several concrete models in spin-$3/2$ system. Our results could help to explore the possible experimental realization of stable time-reversal symmetric topological states.

quant-ph↗

arXiv2021

On Robustness of Neural Semantic Parsers

Semantic parsing maps natural language (NL) utterances into logical forms (LFs), which underpins many advanced NLP problems. Semantic parsers gain performance boosts with deep neural networks, but inherit vulnerabilities against adversarial examples. In this paper, we provide the empirical study on the robustness of semantic parsers in the presence of adversarial attacks. Formally, adversaries of semantic parsing are considered to be the perturbed utterance-LF pairs, whose utterances have exactly the same meanings as the original ones. A scalable methodology is proposed to construct robustness test sets based on existing benchmark corpora. Our results answered five research questions in measuring the sate-of-the-art parsers' performance on robustness test sets, and evaluating the effect of data augmentation.

cs.CL↗

arXiv2020

Stability of Time-Reversal Symmetry Protected Topological Phases

In a closed system, it is well known that the time-reversal symmetry can lead to Kramers degeneracy and protect nontrivial topological states such as quantum spin Hall insulator. In this letter we address the issue whether these effects are stable against coupling to environment, provided that both environment and the coupling to environment also respect the time-reversal symmetry. By employing a non-Hermitian Hamiltonian with the Langevin noise term and ultilizing the non-Hermitian linear response theory, we show that the spectral functions for Kramers degenerate states can be split by dissipation, and the backscattering between counter-propagating edge states can be induced by dissipation. The latter leads to the absence of accurate quantization of conductance in the case of quantum spin Hall effect. As an example, we demonstrate this concretely with the Kane-Mele model. Our study could also be extended to interacting topological phases protected by the time-reversal symmetry.

cond-mat.mes-hall↗

arXiv2020

Security Analysis Methods on Ethereum Smart Contract Vulnerabilities: A Survey

Smart contracts are software programs featuring both traditional applications and distributed data storage on blockchains. Ethereum is a prominent blockchain platform with the support of smart contracts. The smart contracts act as autonomous agents in critical decentralized applications and hold a significant amount of cryptocurrency to perform trusted transactions and agreements. Millions of dollars as part of the assets held by the smart contracts were stolen or frozen through the notorious attacks just between 2016 and 2018, such as the DAO attack, Parity Multi-Sig Wallet attack, and the integer underflow/overflow attacks. These attacks were caused by a combination of technical flaws in designing and implementing software codes. However, many more vulnerabilities of less severity are to be discovered because of the scripting natures of the Solidity language and the non-updateable feature of blockchains. Hence, we surveyed 16 security vulnerabilities in smart contract programs, and some vulnerabilities do not have a proper solution. This survey aims to identify the key vulnerabilities in smart contracts on Ethereum in the perspectives of their internal mechanisms and software security vulnerabilities. By correlating 16 Ethereum vulnerabilities and 19 software security issues, we predict that many attacks are yet to be exploited. And we have explored many software tools to detect the security vulnerabilities of smart contracts in terms of static analysis, dynamic analysis, and formal verification. This survey presents the security problems in smart contracts together with the available analysis tools and the detection methods. We also investigated the limitations of the tools or analysis methods with respect to the identified security vulnerabilities of the smart contracts.

cs.CR↗

arXiv2020

Disconnecting a Traversable Wormhole: Universal Quench Dynamics in Random Spin Models

Understanding strongly interacting quantum matter and quantum gravity are both important open issues in theoretical physics, and the holographic duality between quantum field theory and gravity theory nicely brings these two topics together. Nevertheless, direct connections between gravity physics and experimental observations in quantum matter are still rare. Here we utilize the gravity physics picture to understand quench dynamics experimentally observed in a class of random spin models realized in several different quantum systems, where the dynamics of magnetization are measured after the external polarization field is suddenly turned off. Two universal features of the magnetization dynamics, namely, a slow decay described by a stretched exponential function and an oscillatory behavior, are respectively found in different parameter regimes across different systems. This work addresses the issues of generic conditions under which these two universal features can occur, and we find that a natural answer to this question emerges in the gravity picture. By the holographic duality bridged by a model proposed by Maldacena and Qi, the quench dynamics after suddenly turning off the external polarization field is mapped to disconnecting an eternal traversable wormhole. Our studies show that insight from gravity physics can help unifying different experiments in quantum systems.

cond-mat.quant-gas↗

arXiv2020

Interaction induced dynamical $\mathcal{PT}$ symmetry breaking in dissipative Fermi-Hubbard models

We investigate the dynamical properties of one-dimensional dissipative Fermi-Hubbard models, which are described by the Lindblad master equations with site-dependent jump operators. The corresponding non-Hermitian effective Hamiltonians with pure loss terms possess parity-time ($\mathcal{PT}$) symmetry if we compensate the system additionally an overall gain term. By solving the two-site Lindblad equation with fixed dissipation exactly, we find that the dynamics of rescaled density matrix shows an instability as the interaction increases over a threshold, which can be equivalently described in the scheme of non-Hermitian effective Hamiltonians. This instability is also observed in multi-site systems and closely related to the $\mathcal{PT}$ symmetry breaking accompanied by appearance of complex eigenvalues of the effective Hamiltonian. Moreover, we unveil that the dynamical instability of the anti-ferromagnetic Mott phase comes from the $\mathcal{PT}$ symmetry breaking in highly excited bands, although the low-energy effective model of the non-Hermitian Hubbard model in the strongly interacting regime is always Hermitian. We also provide a quantitative estimation of the time for the observation of dynamical $\mathcal{PT}$ symmetry breaking which could be probed in experiments.

cond-mat.quant-gas↗