Biblio

Stochastic computing is based on probability concepts which are different from conventional mathematical operations. Advantages of stochastic computing in the fields of neural networks and digital image processing have been reported in literature recently. Arithmetic operations especially multiplications can be performed either by logical AND gates in unipolar format or by EXNOR gates in bipolar format in stochastic computation. Stochastic computing is inherently fault-tolerant and requires fewer logic gates to implement arithmetic operations. Long computing time and low accuracy are the main drawbacks of this system. In this presentation, to reduce hardware requirement and delay, modified stochastic multiplication using AND gate array and multiplexer are used for the design of Finite Impulse Response Filter cores. Performance parameters such as area, power and delay for FIR filter using modified stochastic computing methods are compared with conventional floating point computation.

Deep neural networks (DNNs) are effective machine learning models to solve a large class of recognition problems, including the classification of nonlinearly separable patterns. The applications of DNNs are, however, limited by the large size and high energy consumption of the networks. Recently, stochastic computation (SC) has been considered to implement DNNs to reduce the hardware cost. However, it requires a large number of random number generators (RNGs) that lower the energy efficiency of the network. To overcome these limitations, we propose the design of an energy-efficient deep belief network (DBN) based on stochastic computation. An approximate SC activation unit (A-SCAU) is designed to implement different types of activation functions in the neurons. The A-SCAU is immune to signal correlations, so the RNGs can be shared among all neurons in the same layer with no accuracy loss. The area and energy of the proposed design are 5.27% and 3.31% (or 26.55% and 29.89%) of a 32-bit floating-point (or an 8-bit fixed-point) implementation. It is shown that the proposed SC-DBN design achieves a higher classification accuracy compared to the fixed-point implementation. The accuracy is only lower by 0.12% than the floating-point design at a similar computation speed, but with a significantly lower energy consumption.

This paper proposes a game-theoretic approach to analyze the interactions between an attacker and a defender in a cyber-physical system (CPS) and develops effective defense strategies. In a CPS, the attacker launches cyber attacks on a number of nodes in the cyber layer, trying to maximize the potential damage to the underlying physical system while the system operator seeks to defend several nodes in the cyber layer to minimize the physical damage. Given that CPS attacking and defending is often a continual process, a zero-sum Markov game is proposed in this paper to model these interactions subject to underlying uncertainties of real-world events and actions. A novel model is also proposed in this paper to characterize the interdependence between the cyber layer and the physical layer in a CPS and quantify the impact of the cyber attack on the physical damage in the proposed game. To find the Nash equilibrium of the Markov game, we design an efficient algorithm based on value iteration. The proposed general approach is then applied to study the wide-area monitoring and protection issue in smart grid. Extensive simulations are conducted based on real-world data, and results show the effectiveness of the defending strategies derived from the proposed approach.

Popularization of the Internet-of-Things (IoT) has brought widespread concerns on IoT security, especially in face of several recent security incidents related to IoT devices. Due to the resource-constrained nature of many IoT devices, security offloading has been proposed to provide good-enough security for IoT with minimum overhead on the devices. In this paper, we investigate the inevitable risk associated with security offloading: the unprotected and unmonitored transmission from IoT devices to the offloaded security mechanisms. An important challenge in modeling the security risk is the dynamic nature of IoT due to demand fluctuations and infrastructure instability. We propose a stochastic model to capture both the expected and worst-case security risks of an IoT system. We then propose a framework to efficiently address the optimal robust deployment of security mechanisms in IoT. We use results from extensive simulations to demonstrate the superb performance and efficiency of our approach compared to several other algorithms.

On cloud computing systems consisting of task queuing and resource allocations, it is essential but hard to model and evaluate the global performance. In most of the models, researchers use a stochastic process or several stochastic processes to describe a real system. However, due to the absence of theoretical conclusions of any arbitrary stochastic processes, they approximate the complicated model into simple processes that have mathematical results, such as Markov processes. Our purpose is to give a universal method to deal with common stochastic processes as long as the processes can be expressed in the form of transition matrix. To achieve our purpose, we firstly prove several theorems about the convergence of stochastic matrices to figure out what kind of matrix-defined systems has steady states. Furthermore, we propose two strategies for measuring the rate of convergence which reflects how fast the system would come to its steady state. Finally, we give a method for reducing a stochastic matrix into smaller ones, and perform some experiments to illustrate our strategies in practice.

This talk will cover two topics, namely, modeling and design of Moving Target Defense (MTD), and DIFT games for modeling Advanced Persistent Threats (APTs). We will first present a game-theoretic approach to characterizing the trade-off between resource efficiency and defense effectiveness in decoy- and randomization-based MTD. We will then address the game formulation for APTs. APTs are mounted by intelligent and resourceful adversaries who gain access to a targeted system and gather information over an extended period of time. APTs consist of multiple stages, including initial system compromise, privilege escalation, and data exfiltration, each of which involves strategic interaction between the APT and the targeted system. While this interaction can be viewed as a game, the stealthiness, adaptiveness, and unpredictability of APTs imply that the information structure of the game and the strategies of the APT are not readily available. Our approach to modeling APTs is based on the insight that the persistent nature of APTs creates information flows in the system that can be monitored. One monitoring mechanism is Dynamic Information Flow Tracking (DIFT), which taints and tracks malicious information flows through a system and inspects the flows at designated traps. Since tainting all flows in the system will incur significant memory and storage overhead, efficient tagging policies are needed to maximize the probability of detecting the APT while minimizing resource costs. In this work, we develop a multi-stage stochastic game framework for modeling the interaction between an APT and a DIFT, as well as designing an efficient DIFT-based defense. Our model is grounded on APT data gathered using the Refinable Attack Investigation (RAIN) flow-tracking framework. We present the current state of our formulation, insights that it provides on designing effective defenses against APTs, and directions for future work.

Data Stream Machine Learning is rapidly gaining popularity within the network monitoring community as the big data produced by network devices and end-user terminals goes beyond the memory constraints of standard monitoring equipment. Critical network monitoring applications such as the detection of anomalies, network attacks and intrusions, require fast and continuous mechanisms for on-line analysis of data streams. In this paper we consider a stream-based machine learning approach for network security and anomaly detection, applying and evaluating multiple machine learning algorithms in the analysis of continuously evolving network data streams. The continuous evolution of the data stream analysis algorithms coming from the data stream mining domain, as well as the multiple evaluation approaches conceived for benchmarking such kind of algorithms makes it difficult to choose the appropriate machine learning model. Results of the different approaches may significantly differ and it is crucial to determine which approach reflects the algorithm performance the best. We therefore compare and analyze the results from the most recent evaluation approaches for sequential data on commonly used batch-based machine learning algorithms and their corresponding stream-based extensions, for the specific problem of on-line network security and anomaly detection. Similar to our previous findings when dealing with off-line machine learning approaches for network security and anomaly detection, our results suggest that adaptive random forests and stochastic gradient descent models are able to keep up with important concept drifts in the underlying network data streams, by keeping high accuracy with continuous re-training at concept drift detection times.

The paper proposes a new technique to detect Android malware effectively based on converting malware binaries into images and applying machine learning techniques on those images. Existing research converts the whole executable files (e.g., DEX files in Android application package) of target apps into images and uses them for machine learning. However, the entire DEX file (consisting of header section, identifier section, data section, optional link data area, etc.) might contain noisy information for malware detection. In this paper, we convert only data sections of DEX files into grayscale images and apply machine learning on the images with Convolutional Neural Networks (CNN). By using only the data sections for 5,377 malicious and 6,249 benign apps, our technique reduces the storage capacity by 17.5% on average compared to using the whole DEX files. We apply two CNN models, Inception-v3 and Inception-ResNet-v2, which are known to be efficient in image processing, and examine the effectiveness of our technique in terms of accuracy. Experiment results show that the proposed technique achieves better accuracy with smaller storage capacity than the approach using the whole DEX files. Inception-ResNet-v2 with the stochastic gradient descent (SGD) optimization algorithm reaches 98.02% accuracy.

Many applications, such as anonymous communication systems, privacy-enhancing database queries, or privacy-enhancing machine-learning methods, require robust guarantees under thousands and sometimes millions of observations. The notion of r-fold approximate differential privacy (ADP) offers a well-established framework with a precise characterization of the degree of privacy after r observations of an attacker. However, existing bounds for r-fold ADP are loose and, if used for estimating the required degree of noise for an application, can lead to over-cautious choices for perturbation randomness and thus to suboptimal utility or overly high costs. We present a numerical and widely applicable method for capturing the privacy loss of differentially private mechanisms under composition, which we call privacy buckets. With privacy buckets we compute provable upper and lower bounds for ADP for a given number of observations. We compare our bounds with state-of-the-art bounds for r-fold ADP, including Kairouz, Oh, and Viswanath's composition theorem (KOV), concentrated differential privacy and the moments accountant. While KOV proved optimal bounds for heterogeneous adaptive k-fold composition, we show that for concrete sequences of mechanisms tighter bounds can be derived by taking the mechanisms' structure into account. We compare previous bounds for the Laplace mechanism, the Gauss mechanism, for a timing leakage reduction mechanism, and for the stochastic gradient descent and we significantly improve over their results (except that we match the KOV bound for the Laplace mechanism, for which it seems tight). Our lower bounds almost meet our upper bounds, showing that no significantly tighter bounds are possible.

Probabilistic model checking is a formal verification technique that has been applied successfully in a variety of domains, providing identification of system errors through quantitative verification of stochastic system models. One domain that can benefit from probabilistic model checking is cloud computing, which must provide highly reliable and secure computational and storage services to large numbers of mission-critical software systems. For real-world domains like cloud computing, external system factors and environmental changes must be estimated accurately in the form of probabilities in system models; inaccurate estimates for the model probabilities can lead to invalid verification results. To address the effects of uncertainty in probability estimates, in previous work we have developed a variety of techniques for perturbation analysis of discrete- and continuous-time Markov chains (DTMCs and CTMCs). These techniques determine the consequences of the uncertainty on verification of system properties. In this paper, we present the first approach for perturbation analysis of Markov decision processes (MDPs), a stochastic formalism that is especially popular due to the significant expressive power it provides through the combination of both probabilistic and nondeterministic choice. Our primary contribution is a novel technique for efficiently analyzing the effects of perturbations of model probabilities on verification of reachability properties of MDPs. The technique heuristically explores the space of adversaries of an MDP, which encode the different ways of resolving the MDP’s nondeterministic choices. We demonstrate the practical effectiveness of our approach by applying it to two case studies of cloud systems.

While significant progress has been made separately on analytics systems for scalable stochastic gradient descent (SGD) and private SGD, none of the major scalable analytics frameworks have incorporated differentially private SGD. There are two inter-related issues for this disconnect between research and practice: (1) low model accuracy due to added noise to guarantee privacy, and (2) high development and runtime overhead of the private algorithms. This paper takes a first step to remedy this disconnect and proposes a private SGD algorithm to address both issues in an integrated manner. In contrast to the white-box approach adopted by previous work, we revisit and use the classical technique of output perturbation to devise a novel “bolt-on” approach to private SGD. While our approach trivially addresses (2), it makes (1) even more challenging. We address this challenge by providing a novel analysis of the L2-sensitivity of SGD, which allows, under the same privacy guarantees, better convergence of SGD when only a constant number of passes can be made over the data. We integrate our algorithm, as well as other state-of-the-art differentially private SGD, into Bismarck, a popular scalable SGD-based analytics system on top of an RDBMS. Extensive experiments show that our algorithm can be easily integrated, incurs virtually no overhead, scales well, and most importantly, yields substantially better (up to 4X) test accuracy than the state-of-the-art algorithms on many real datasets.

In many Internet of Thing (IoT) application domains security is a critical requirement, because malicious parties can undermine the effectiveness of IoT-based systems by compromising single components and/or communication channels. Thus, a security infrastructure is needed to ensure the proper functioning of such systems even under attack. However, it is also critical that security be at a reasonable resource and energy cost, as many IoT devices may not have sufficient resources to host expensive security tools. In this paper, we focus on the problem of efficiently and effectively securing IoT networks by carefully allocating security tools. We model our problem according to game theory, and provide a Pareto-optimal solution, in which the cost of the security infrastructure, its energy consumption, and the probability of a successful attack, are minimized. Our experimental evaluation shows that our technique improves the system robustness in terms of packet delivery rate for different network topologies.

Cloud Data Center (CDC) security remains a major challenge for business organizations and takes an important concern with research works. The attacker purpose is to guarantee the service unavailability and maximize the financial loss costs. As a result, Distributed Denial of Service (DDoS) attacks have appeared as the most popular attack. The main aim of such attacks is to saturate and overload the system network through a massive data packets size flooding toward a victim server and to block the service to users. This paper provides a defending system in order to mitigate the Denial of Service (DoS) attack in CDC environment. Basically it outlines the different techniques of DoS attacks and its countermeasures by combining the filtering and detection mechanisms. We presented an analytical model based on queueing model to evaluate the impact of flooding attack on cloud environment regarding service availability and QoS performance. Consequently, we have plotted the response time, throughput, drop rate and resource computing utilization varying the attack arrival rate. We have used JMT (Java Modeling Tool) simulator to validate the analytical model. Our approach was appeared powerful for attacks mitigation in the cloud environment.

Intrusion detection systems do not perform well when it comes to detecting zero-day attacks, therefore improving their performance in that regard is an active research topic. In this study, to detect zero-day attacks with high accuracy, we proposed two deep learning based anomaly detection models using autoencoder and denoising autoencoder respectively. The key factor that directly affects the accuracy of the proposed models is the threshold value which was determined using a stochastic approach rather than the approaches available in the current literature. The proposed models were tested using the KDDTest+ dataset contained in NSL-KDD, and we achieved an accuracy of 88.28% and 88.65% respectively. The obtained results show that, as a singular model, our proposed anomaly detection models outperform any other singular anomaly detection methods and they perform almost the same as the newly suggested hybrid anomaly detection models.

Software development for Cyber-Physical Systems (CPS), e.g., autonomous vehicles, requires both functional and non-functional quality assurance to guarantee that the CPS operates safely and effectively. EAST-ADL is a domain specific architectural language dedicated to safety-critical automotive embedded system design. We have previously modified EAST-ADL to include energy constraints and transformed energy-aware real-time (ERT) behaviors modeled in EAST-ADL/Stateflow into UPPAAL models amenable to formal verification. Previous work is extended in this paper by including support for Simulink and an integration of Simulink/Stateflow (S/S) within the same too lchain. S/S models are transformed, based on the extended ERT constraints with probability parameters, into verifiable UPPAAL-SMC models and integrate the translation with formal statistical analysis techniques: Probabilistic extension of EAST-ADL constraints is defined as a semantics denotation. A set of mapping rules is proposed to facilitate the guarantee of translation. Formal analysis on both functional- and non-functional properties is performed using Simulink Design Verifier and UPPAAL-SMC. Our approach is demonstrated on the autonomous traffic sign recognition vehicle case study.

We present a formal method for computing the best security provisioning for Internet of Things (IoT) scenarios characterized by a high degree of mobility. The security infrastructure is intended as a security resource allocation plan, computed as the solution of an optimization problem that minimizes the risk of having IoT devices not monitored by any resource. We employ the shortfall as a risk measure, a concept mostly used in the economics, and adapt it to our scenario. We show how to compute and evaluate an allocation plan, and how such security solutions address the continuous topology changes that affect an IoT environment.

In this study, it is proposed to carry out an efficient formulation in order to figure out the stochastic security-constrained generation capacity expansion planning (SC-GCEP) problem. The main idea is related to directly compute the line outage distribution factors (LODF) which could be applied to model the N - m post-contingency analysis. In addition, the post-contingency power flows are modeled based on the LODF and the partial transmission distribution factors (PTDF). The post-contingency constraints have been reformulated using linear distribution factors (PTDF and LODF) so that both the pre- and post-contingency constraints are modeled simultaneously in the SC-GCEP problem using these factors. In the stochastic formulation, the load uncertainty is incorporated employing a two-stage multi-period framework, and a K - means clustering technique is implemented to decrease the number of load scenarios. The main advantage of this methodology is the feasibility to quickly compute the post-contingency factors especially with multiple-line outages (N - m). This concept would improve the security-constraint analysis modeling quickly the outage of m transmission lines in the stochastic SC-GCEP problem. It is carried out several experiments using two electrical power systems in order to validate the performance of the proposed formulation.

This study proposes to apply an efficient formulation to solve the stochastic security-constrained generation capacity expansion planning (GCEP) problem using an improved method to directly compute the generalized generation distribution factors (GGDF) and the line outage distribution factors (LODF) in order to model the pre- and the post-contingency constraints based on the only application of the partial transmission distribution factors (PTDF). The classical DC-based formulation has been reformulated in order to include the security criteria solving both pre- and post-contingency constraints simultaneously. The methodology also takes into account the load uncertainty in the optimization problem using a two-stage multi-period model, and a clustering technique is used as well to reduce load scenarios (stochastic problem). The main advantage of this methodology is the feasibility to quickly compute the LODF especially with multiple-line outages (N-m). This idea could speed up contingency analyses and improve significantly the security-constrained analyses applied to GCEP problems. It is worth to mentioning that this approach is carried out without sacrificing optimality.

Stochastic Computing (SC) is an alternative design paradigm particularly useful for applications where cost is critical. SC has been applied to neural networks, as neural networks are known for their high computational complexity. However previous work in this area has critical limitations such as the fully-parallel architecture assumption, which prevent them from being applicable to recent ones such as convolutional neural networks, or ConvNets. This paper presents the first SC architecture for ConvNets, shows its feasibility, with detailed analyses of implementation overheads. Our SC-ConvNet is a hybrid between SC and conventional binary design, which is a marked difference from earlier SC-based neural networks. Though this might seem like a compromise, it is a novel feature driven by the need to support modern ConvNets at scale, which commonly have many, large layers. Our proposed architecture also features hybrid layer composition, which helps achieve very high recognition accuracy. Our detailed evaluation results involving functional simulation and RTL synthesis suggest that SC-ConvNets are indeed competitive with conventional binary designs, even without considering inherent error resilience of SC.