arXiv subjects

Jingyi Wang

Publications and source records attributed to Jingyi Wang.

At least 145 records · Page 8Linked to original sources

arXiv2021

RobOT: Robustness-Oriented Testing for Deep Learning Systems

Recently, there has been a significant growth of interest in applying software engineering techniques for the quality assurance of deep learning (DL) systems. One popular direction is deep learning testing, where adversarial examples (a.k.a.~bugs) of DL systems are found either by fuzzing or guided search with the help of certain testing metrics. However, recent studies have revealed that the commonly used neuron coverage metrics by existing DL testing approaches are not correlated to model robustness. It is also not an effective measurement on the confidence of the model robustness after testing. In this work, we address this gap by proposing a novel testing framework called Robustness-Oriented Testing (RobOT). A key part of RobOT is a quantitative measurement on 1) the value of each test case in improving model robustness (often via retraining), and 2) the convergence quality of the model robustness improvement. RobOT utilizes the proposed metric to automatically generate test cases valuable for improving model robustness. The proposed metric is also a strong indicator on how well robustness improvement has converged through testing. Experiments on multiple benchmark datasets confirm the effectiveness and efficiency of RobOT in improving DL model robustness, with 67.02% increase on the adversarial robustness that is 50.65% higher than the state-of-the-art work DeepGini.

cs.SE↗

arXiv2020

Improving Neural Network Verification through Spurious Region Guided Refinement

We propose a spurious region guided refinement approach for robustness verification of deep neural networks. Our method starts with applying the DeepPoly abstract domain to analyze the network. If the robustness property cannot be verified, the result is inconclusive. Due to the over-approximation, the computed region in the abstraction may be spurious in the sense that it does not contain any true counterexample. Our goal is to identify such spurious regions and use them to guide the abstraction refinement. The core idea is to make use of the obtained constraints of the abstraction to infer new bounds for the neurons. This is achieved by linear programming techniques. With the new bounds, we iteratively apply DeepPoly, aiming to eliminate spurious regions. We have implemented our approach in a prototypical tool DeepSRGR. Experimental results show that a large amount of regions can be identified as spurious, and as a result, the precision of DeepPoly can be significantly improved. As a side contribution, we show that our approach can be applied to verify quantitative robustness properties.

cs.AI↗

arXiv2020

Towards Interpreting Recurrent Neural Networks through Probabilistic Abstraction

Neural networks are becoming a popular tool for solving many real-world problems such as object recognition and machine translation, thanks to its exceptional performance as an end-to-end solution. However, neural networks are complex black-box models, which hinders humans from interpreting and consequently trusting them in making critical decisions. Towards interpreting neural networks, several approaches have been proposed to extract simple deterministic models from neural networks. The results are not encouraging (e.g., low accuracy and limited scalability), fundamentally due to the limited expressiveness of such simple models. In this work, we propose an approach to extract probabilistic automata for interpreting an important class of neural networks, i.e., recurrent neural networks. Our work distinguishes itself from existing approaches in two important ways. One is that probability is used to compensate for the loss of expressiveness. This is inspired by the observation that human reasoning is often `probabilistic'. The other is that we adaptively identify the right level of abstraction so that a simple model is extracted in a request-specific way. We conduct experiments on several real-world datasets using state-of-the-art architectures including GRU and LSTM. The result shows that our approach significantly improves existing approaches in terms of accuracy or scalability. Lastly, we demonstrate the usefulness of the extracted models through detecting adversarial texts.

cs.LG↗

arXiv2020

Towards Plausible Differentially Private ADMM Based Distributed Machine Learning

The Alternating Direction Method of Multipliers (ADMM) and its distributed version have been widely used in machine learning. In the iterations of ADMM, model updates using local private data and model exchanges among agents impose critical privacy concerns. Despite some pioneering works to relieve such concerns, differentially private ADMM still confronts many research challenges. For example, the guarantee of differential privacy (DP) relies on the premise that the optimality of each local problem can be perfectly attained in each ADMM iteration, which may never happen in practice. The model trained by DP ADMM may have low prediction accuracy. In this paper, we address these concerns by proposing a novel (Improved) Plausible differentially Private ADMM algorithm, called PP-ADMM and IPP-ADMM. In PP-ADMM, each agent approximately solves a perturbed optimization problem that is formulated from its local private data in an iteration, and then perturbs the approximate solution with Gaussian noise to provide the DP guarantee. To further improve the model accuracy and convergence, an improved version IPP-ADMM adopts sparse vector technique (SVT) to determine if an agent should update its neighbors with the current perturbed solution. The agent calculates the difference of the current solution from that in the last iteration, and if the difference is larger than a threshold, it passes the solution to neighbors; or otherwise the solution will be discarded. Moreover, we propose to track the total privacy loss under the zero-concentrated DP (zCDP) and provide a generalization performance analysis. Experiments on real-world datasets demonstrate that under the same privacy guarantee, the proposed algorithms are superior to the state of the art in terms of model accuracy and convergence rate.

cs.LG↗

arXiv2020

Relativistic reflection and reverberation in GX 339-4 with NICER and NuSTAR

We analyze seven NICER and NuSTAR epochs of the black hole X-ray binary GX 339-4 in the hard state during its two most recent hard-only outbursts in 2017 and 2019. These observations cover the 1-100 keV unabsorbed luminosities between 0.3% and 2.1% of the Eddington limit. With NICER's negligible pile-up, high count rate and unprecedented time resolution, we perform a spectral-timing analysis and spectral modeling using relativistic and distant reflection models. Our spectral fitting shows that as the inner disk radius moves inwards, the thermal disk emission increases in flux and temperature, the disk becomes more highly ionized and the reflection fraction increases. This coincides with the inner disk increasing its radiative efficiency around ~1% Eddington. We see a hint of hysteresis effect at ~0.3% of Eddington: the inner radius is significantly truncated during the rise ($>49R_{g}$), while only a mild truncation ($\sim5R_g$) is found during the decay. At higher frequencies ($2-7$~Hz) in the highest luminosity epoch, a soft lag is present, whose energy dependence reveals a thermal reverberation lag, with an amplitude similar to previous findings for this source. We also discuss the plausibility of the hysteresis effect and the debate of the disk truncation problem in the hard state.

astro-ph.HE↗

arXiv2020

Testing the Kerr metric using X-ray reflection spectroscopy: spectral analysis of GX 339-4

Signatures of X-ray reprocessing (reflection) out of an accretion disk are commonly observed in the high-energy spectrum of accreting black holes, and can be used to probe the strong gravity region around these objects. In this paper, we extend previous work in the literature and we employ a full emission model for relativistic reflection in non-Kerr spacetime to demonstrate an approach that tests the Kerr black hole hypothesis. We analyze a composite spectrum obtained with the Proportional Counter Array in the Rossi X-ray Timing Explorer (RXTE), of the stellar-mass black hole GX 339-4 in its brightest hard state. With a remarkable sensitivity of ~0.1% and 40 million counts in the 3-45 keV band to capture the faint features in the reflection spectrum, we demonstrate that it is possible with existing data and an adequate model to place constraints on the black hole spin $a_*$ and the deformation parameter that quantifies the departure from the Kerr metric. Our measurement obtained with the best fit model, which should be regarded as principally a proof of concept, is $a_*=0.92^{+0.07}_{-0.12}$ and $α_{13}=-0.76^{+0.78}_{-0.60}$ with a 90% confidence level and is consistent with the hypothesis that the compact object in GX 339-4 is a Kerr black hole. We also discuss how the physical model choice and the emissivity profile adopted could make an impact on the constraints of $α_{13}$ and spin. To enable Kerr metric test using X-ray reflection spectroscopy, it is essential to improve our astrophysical understanding of accreting black holes, e.g., the natures of accretion flow and corona.

gr-qc↗

arXiv2019

Water Supply Prediction Based on Initialized Attention Residual Network

Real-time and accurate water supply forecast is crucial for water plant. However, most existing methods are likely affected by factors such as weather and holidays, which lead to a decline in the reliability of water supply prediction. In this paper, we address a generic artificial neural network, called Initialized Attention Residual Network (IARN), which is combined with an attention module and residual modules. Specifically, instead of continuing to use the recurrent neural network (RNN) in time-series tasks, we try to build a convolution neural network (CNN)to recede the disturb from other factors, relieve the limitation of memory size and get a more credible results. Our method achieves state-of-the-art performance on several data sets, in terms of accuracy, robustness and generalization ability.

cs.LG↗

arXiv2019

There is Limited Correlation between Coverage and Robustness for Deep Neural Networks

Deep neural networks (DNN) are increasingly applied in safety-critical systems, e.g., for face recognition, autonomous car control and malware detection. It is also shown that DNNs are subject to attacks such as adversarial perturbation and thus must be properly tested. Many coverage criteria for DNN since have been proposed, inspired by the success of code coverage criteria for software programs. The expectation is that if a DNN is a well tested (and retrained) according to such coverage criteria, it is more likely to be robust. In this work, we conduct an empirical study to evaluate the relationship between coverage, robustness and attack/defense metrics for DNN. Our study is the largest to date and systematically done based on 100 DNN models and 25 metrics. One of our findings is that there is limited correlation between coverage and robustness, i.e., improving coverage does not help improve the robustness. Our dataset and implementation have been made available to serve as a benchmark for future studies on testing DNN.

cs.LG↗

arXiv2019

The 2017 Failed Outburst of GX 339-4: Relativistic X-ray Reflection near the Black Hole Revealed by NuSTAR and Swift Spectroscopy

We report on the spectroscopic analysis of the black hole binary GX 339-4 during its recent 2017-2018 outburst, observed simultaneously by the Swift and NuSTAR observatories. Although during this particular outburst the source failed to make state transitions, and despite Sun constraints during the peak luminosity, we were able to trigger four different observations sampling the evolution of the source in the hard state. We show that even for the lowest luminosity observations the NuSTAR spectra show clear signatures of X-ray reprocessing (reflection) in an accretion disk. Detailed analysis of the highest signal-to-noise spectra with our family of relativistic reflection models RELXILL indicates the presence of both broad and narrow reflection components. We find that a dual-lamppost model provides a superior fit when compared to the standard single lamppost plus distant neutral reflection. In the dual lamppost model two sources at different heights are placed on the rotational axis of the black hole, suggesting that the narrow component of the Fe K emission is likely to originate in regions far away in the disk, but still significantly affected by its rotational motions. Regardless of the geometry assumed, we find that the inner edge of the accretion disk reaches a few gravitational radii in all our fits, consistent with previous determinations at similar luminosity levels. This confirms a very low degree of disk truncation for this source at luminosities above ~1% Eddington. Our estimates of Rin reinforces the suggested behavior for an inner disk that approaches the inner-most regions as the luminosity increases in the hard state.

astro-ph.HE↗

arXiv2019

Synchronization of Networked Harmonic Oscillators via Quantized Sampled Velocity Feedback

In this technical note, we propose a practicable quantized sampled velocity data coupling protocol for synchronization of a set of harmonic oscillators. The coupling protocol is designed in a quantized way via interconnecting the velocities encoded by a uniform quantizer with a zooming parameter in either a fixed or an adjustable form over a directed communication network. We establish sufficient conditions for the networked harmonic oscillators to converge to a bounded neighborhood of the synchronized orbits with a fixed zooming parameter. We ensure the oscillators to achieve synchronization by designing the quantized coupling protocol with an adjustable zooming parameter. Finally, we show two numerical examples to illustrate the effectiveness of the proposed coupling protocol.

math.OC↗

arXiv2019

Adversarial Sample Detection for Deep Neural Network through Model Mutation Testing

Deep neural networks (DNN) have been shown to be useful in a wide range of applications. However, they are also known to be vulnerable to adversarial samples. By transforming a normal sample with some carefully crafted human imperceptible perturbations, even highly accurate DNN make wrong decisions. Multiple defense mechanisms have been proposed which aim to hinder the generation of such adversarial samples. However, a recent work show that most of them are ineffective. In this work, we propose an alternative approach to detect adversarial samples at runtime. Our main observation is that adversarial samples are much more sensitive than normal samples if we impose random mutations on the DNN. We thus first propose a measure of `sensitivity' and show empirically that normal samples and adversarial samples have distinguishable sensitivity. We then integrate statistical hypothesis testing and model mutation testing to check whether an input sample is likely to be normal or adversarial at runtime by measuring its sensitivity. We evaluated our approach on the MNIST and CIFAR10 datasets. The results show that our approach detects adversarial samples generated by state-of-the-art attacking methods efficiently and accurately.

cs.LG↗

arXiv2019

High Density Reflection Spectroscopy I. A case study of GX~339-4

We present a broad band spectral analysis of the black hole binary GX~339-4 with NuSTAR and Swift using high density reflection model. The observations were taken when the source was in low flux hard states (LF) during the outbursts in 2013 and 2015, and in a very high flux soft state (HF) in 2015. The high density reflection model can explain its LF spectra with no requirement for an additional low temperature thermal component. This model enables us to constrain the density in the disc surface of GX~339-4 in different flux states. The disc density in the LF state is $\log(n_{\rm e}/$ cm$^{-3})\approx21$, 100 times higher than the density in the HF state ($\log(n_{\rm e}/$ cm$^{-3})=18.93^{+0.12}_{-0.16}$). A close-to-solar iron abundance is obtained by modelling the LF and HF broad band spectra with variable density reflection model ($Z_{\rm Fe}=1.50^{+0.12}_{-0.04}Z_{\odot}$ and $Z_{\rm Fe}=1.05^{+0.17}_{-0.15}Z_{\odot}$ respectively).

astro-ph.HE↗

arXiv2018

Automatically 'Verifying' Discrete-Time Complex Systems through Learning, Abstraction and Refinement

Precisely modeling complex systems like cyber-physical systems is challenging, which often render model-based system verification techniques like model checking infeasible. To overcome this challenge, we propose a method called LAR to automatically `verify' such complex systems through a combination of learning, abstraction and refinement from a set of system log traces. We assume that log traces and sampling frequency are adequate to capture `enough' behaviour of the system. Given a safety property and the concrete system log traces as input, LAR automatically learns and refines system models, and produces two kinds of outputs. One is a counterexample with a bounded probability of being spurious. The other is a probabilistic model based on which the given property is `verified'. The model can be viewed as a proof obligation, i.e., the property is verified if the model is correct. It can also be used for subsequent system analysis activities like runtime monitoring or model-based testing. Our method has been implemented as a self-contained software toolkit. The evaluation on multiple benchmark systems as well as a real-world water treatment system shows promising results.

cs.SE↗

arXiv2018

Detecting Adversarial Samples for Deep Neural Networks through Mutation Testing

Recently, it has been shown that deep neural networks (DNN) are subject to attacks through adversarial samples. Adversarial samples are often crafted through adversarial perturbation, i.e., manipulating the original sample with minor modifications so that the DNN model labels the sample incorrectly. Given that it is almost impossible to train perfect DNN, adversarial samples are shown to be easy to generate. As DNN are increasingly used in safety-critical systems like autonomous cars, it is crucial to develop techniques for defending such attacks. Existing defense mechanisms which aim to make adversarial perturbation challenging have been shown to be ineffective. In this work, we propose an alternative approach. We first observe that adversarial samples are much more sensitive to perturbations than normal samples. That is, if we impose random perturbations on a normal and an adversarial sample respectively, there is a significant difference between the ratio of label change due to the perturbations. Observing this, we design a statistical adversary detection algorithm called nMutant (inspired by mutation testing from software engineering community). Our experiments show that nMutant effectively detects most of the adversarial samples generated by recently proposed attacking methods. Furthermore, we provide an error bound with certain statistical significance along with the detection.

cs.LG↗

arXiv2018

Toward `verifying' a Water Treatment System

Modeling and verifying real-world cyber-physical systems is challenging, which is especially so for complex systems where manually modeling is infeasible. In this work, we report our experience on combining model learning and abstraction refinement to analyze a challenging system, i.e., a real-world Secure Water Treatment system (SWaT). Given a set of safety requirements, the objective is to either show that the system is safe with a high probability (so that a system shutdown is rarely triggered due to safety violation) or not. As the system is too complicated to be manually modeled, we apply latest automatic model learning techniques to construct a set of Markov chains through abstraction and refinement, based on two long system execution logs (one for training and the other for testing). For each probabilistic safety property, we either report it does not hold with a certain level of probabilistic confidence, or report that it holds by showing the evidence in the form of an abstract Markov chain. The Markov chains can subsequently be implemented as runtime monitors in SWaT.

cs.AI↗

arXiv2017

Aiding autobiographic memory by using wearable devices

In this paper, we investigate the effectiveness of two distinct techniques (Special Moment Approach & Spatial Frequency Approach) for reviewing the lifelogs, which were collected by lifeloggers who were willing to use a wearable camera and a bracelet simultaneously for two days. Generally, Special moment approach is a technique for extracting episodic events and Spatial frequency approach is a technique for associating visual with temporal and location information, especially heat map is applied as the spatial data for expressing frequency awareness. Based on that, the participants were asked to fill in two post-study questionnaires for evaluating the effectiveness of those two techniques and their combination. The preliminary result showed the positive potential of exploring individual lifelogs using our approaches.

cs.HC↗

arXiv2017

Should We Learn Probabilistic Models for Model Checking? A New Approach and An Empirical Study

Many automated system analysis techniques (e.g., model checking, model-based testing) rely on first obtaining a model of the system under analysis. System modeling is often done manually, which is often considered as a hindrance to adopt model-based system analysis and development techniques. To overcome this problem, researchers have proposed to automatically "learn" models based on sample system executions and shown that the learned models can be useful sometimes. There are however many questions to be answered. For instance, how much shall we generalize from the observed samples and how fast would learning converge? Or, would the analysis result based on the learned model be more accurate than the estimation we could have obtained by sampling many system executions within the same amount of time? In this work, we investigate existing algorithms for learning probabilistic models for model checking, propose an evolution-based approach for better controlling the degree of generalization and conduct an empirical study in order to answer the questions. One of our findings is that the effectiveness of learning may sometimes be limited.

cs.SE↗

arXiv2016

Learning Likely Invariants to Explain Why a Program Fails

Debugging is difficult. Recent studies show that automatic bug localization techniques have limited usefulness. One of the reasons is that programmers typically have to understand why the program fails before fixing it. In this work, we aim to help programmers understand a bug by automatically generating likely invariants which are violated in the failed tests. Given a program with an initial assertion and at least one test case failing the assertion, we first generate random test cases, identify potential bug locations through bug localization, and then generate program state mutation based on active learning techniques to identify a predicate "explaining" the cause of the bug. The predicate is a classifier for the passed test cases and failed test cases. Our main contribution is the application of invariant learning for bug explanation, as well as a novel approach to overcome the problem of lack of test cases in practice. We apply our method to real-world bugs and show the generated invariants are often correlated to the actual bug fixes.

cs.SE↗