Differentially Private Kernel Support Vector Machines Based on the Exponential and Laplace Hybrid Mechanism

Support vector machines (SVMs) are among the most robust and accurate methods in all well-known machine learning algorithms, especially for classification. The SVMs train a classification model by solving an optimization problem to decide which instances in the training datasets are the support vectors (SVs). However, SVs are intact instances taken from the training datasets and directly releasing the classification model of the SVMs will carry significant risk to the privacy of individuals, when the training datasets contain sensitive information. In this paper, we study the problem of how to release the classification model of kernel SVMs while preventing privacy leakage of the SVs and satisfying the requirement of privacy protection. We propose a new differentially private algorithm for the kernel SVMs based on the exponential and Laplace hybrid mechanism named DPKSVMEL. The DPKSVMEL algorithm has two major advantages compared with existing private SVM algorithms. One is that it protects the privacy of the SVs by postprocessing and the training process of the non-private kernel SVMs does not change. Another is that the scoring function values are directly derived from the symmetric kernel matrix generated during the training process and does not require additional storage space and complex sensitivity analysis. In the DPKSVMEL algorithm, we define a similarity parameter to denote the correlation or distance between the non-SVs and every SV. And then, every non-SV is divided into a group with one of the SVs according to the maximal value of the similarity. Under some certain similarity parameter value, we replace every SV with a mean value of the top-k randomly selected most similar non-SVs within the group by the exponential mechanism if the number of non-SVs is greater than k. Otherwise, we add random noise to the SVs by the Laplace mechanism. We theoretically prove that the DPKSVMEL algorithm satisfies differential privacy. The extensive experiments show the effectiveness of the DPKSVMEL algorithm for kernel SVMs on real datasets; meanwhile, it achieves higher classification accuracy than existing private SVM algorithms.

Numerical Modeling of Failure Mechanisms in Articulated Concrete Block Mattress as a Sustainable Coastal Protection Structure

Shoreline protection remains a global priority. Typically, coastal areas are protected by armoring them with hard, non-native, and non-sustainable materials such as limestone. To increase the execution speed and environmental friendliness and reduce the weight of individual concrete blocks and reinforcements, concrete blocks can be designed and implemented as Articulated Concrete Block Mattress (ACB Mat). These structures act as an integral part and can be used as a revetment on the breakwater body or shoreline protection. Physical models are one of the key tools for estimating and investigating the phenomena in coastal structures. However, it does have limitations and obstacles; consequently, in this study, numerical modeling of waves on these structures has been utilized to simulate wave propagation on the breakwater, via Flow-3D software with VOF. Among the factors affecting the instability of ACB Mat are breaking waves as well as the shaking of the revetment and the displacement of the armor due to the uplift force resulting from the failure. The most important purpose of the present study is to investigate the ability of numerical Flow-3D model to simulate hydrodynamic parameters in coastal revetment. The run-up values of the waves on the concrete block armoring will multiply with increasing break parameter (0.5<ξm−1,0<3.3) due to the existence of plunging waves until it (Ru2%Hm0=1.6) reaches maximum. Hence, by increasing the breaker parameter and changing breaking waves (ξm−1,0>3.3) type to collapsing waves/surging waves, the trend of relative wave run-up changes on concrete block revetment increases gradually. By increasing the breaker index (surf similarity parameter) in the case of plunging waves (0.5<ξm−1,0<3.3), the low values on the relative wave run-down are greatly reduced. Additionally, in the transition region, the change of breaking waves from plunging waves to collapsing/surging (3.3<ξm−1,0<5.0), the relative run-down process occurs with less intensity.

Self-Similarity in Magnetostrictive Materials: An Experimental Point of View

Magnetostrictive behavior is characterized by a complex coupling between magnetic and mechanical quantities. While this behavior can be quite easily exploited for both actuation and sensing or energy conversion purposes, the complex hysteresis interaction between magnetization and magnetic field and mechanical stress and strain is hard to model. Nevertheless, magnetic and magnetostrictive experimental curves are quite self-similar, assuming stress as self-similarity parameter. The quantification of this concept would help modeling. Here, this concept is quantified and experimentally confirmed over different types of magnetostrictive samples.

Combined Longshore and Cross-Shore Modeling for Low-Energy Embayed Sandy Beaches

The present study focuses on the long-term multi-year evolution of the shoreline position of the Nha Trang sandy beach. To this end an empirical model which is a combination of longshore and cross-shore models, is used. The Nha Trang beach morphology is driven by a tropical wave climate dominated by seasonal variations and winter monsoon intra-seasonal pulses. The combined model accounts for seasonal shoreline evolution, which is primarily attributed to cross-shore dynamics but fails to represent accretion that occurs during the height of summer under low energy conditions. The reason is in the single equilibrium Dean number Ωeq of the ShoreFor model, one of the components of the combined model. This equilibrium Dean number cannot simultaneously account for the evolution of strong intra-seasonal events (i.e., winter monsoon pulses) and the annual recovery mechanisms associated with swash transport. By assigning a constant value to Ωeq, when the surf similarity parameter is higher than 3.3 (occurrence of small surging breakers in summer), we strongly improve the shoreline position prediction. This clearly points to the relevance of a multi-scale approach, although our modified Ωeq retains the advantage of simplicity.

Field Observations of a Multilevel Beach Cusp System and Their Swash Zone Dynamics

This paper presents the observed morphological evolution of a multilevel beach cusp system in Long Strand, Co. Cork, Ireland. The surveys were carried out with an Unmanned Aerial Vehicle (UAV) system between March and September 2019. From this site, three levels of beach cusps on the beachface (i.e., lower beach level, mid beach level and upper beach level), and critical cusp parameters are reported, including cusp spacing, cusp elevation, cusp depth, and cusp amplitude. Thus far, such an extensive dataset has not previously been reported in the literature from a single site. The evolution of the different cusp parameters is then linked with the hydrodynamics in the study area, and new prediction theories are proposed for the different cusp parameters. The Lower beach level cusps (1 < z < 2.5 m Irish Transverse Mercator (ITM)) changed with every tide and appeared when surf-similarity parameter -ξ0 < 1.55. These cusps had a mean cusp spacing of λmean = 11.09 m, which are closely linked with the predictions of the self-organisation theory (p < 0.05). In contrast, the Mid beach level cusps (2.5 < z < 3.5 m ITM) are less dynamic compared to the Lower beach level cusps and can persist between spring tidal cycles. They had a mean cusp spacing of λmean = 18.17 m. The Upper beach level cusps (approximately z = 6 m ITM) are above astronomical tide levels and have a mean cusp spacing of λmean = 40.26 m. They did not change significantly over the survey period due to a lack of major storm events. These findings give a better understanding of the evolution of different cusp parameters for a multilevel beach cusp system and can be used to formulate a global theory regarding their change over time.

Detection of malware using an artificial neural network based on adaptive resonant theory

The process of detecting malicious code by anti-virus systems is considered. The main part of this process is the procedure for analyzing a file or process. Artificial neural networks based on the adaptive-resonance theory are proposed to use as a method of analysis. The graph2vec vectorization algorithm is used to represent the analyzed program codes in numerical format. Despite the fact that the use of this vectorization method ignores the semantic relationships between the sequence of executable commands, it allows to reduce the analysis time without significant loss of accuracy. The use of an artificial neural network ART-2m with a hierarchical memory structure made it possible to reduce the classification time for a malicious file. Reducing the classification time allows to set more memory levels and increase the similarity parameter, which leads to an improved classification quality. Experiments show that with this approach to detecting malicious software, similar files can be recognized by both size and behavior.

Statistical research and modeling network traffic

The self-similarity properties of the considered traffic were checked on different time scales obtained on the available daily traffic data. An estimate of the tail severity of the distribution self-similar traffic was obtained by constructing a regression line for the additional distribution function on a logarithmic scale. The self-similarity parameter value, determined by the severity of the distribution “tail”, made it possible to confirm the assumption of traffic self-similarity. A review of models simulating real network traffic with a self-similar structure was made. Implemented tools for generating artificial traffic in accordance with the considered models. Made comparison of artificial network traffic generators according to the least squares method criterion for approximating the artificial traffic point values by the approximation function of traffic. Qualitative assessments traffic generators in the form of the software implementation complexity were taken into account, which, however, can be a subjective assessment. Comparative characteristics allow you to choose some generators that most faithfully simulate real network traffic. The proposed sequence of methods to study the network traffic properties is necessary to understand its nature and to develop appropriate models that simulate real network traffic.

Superheating Limit of Hydrocarbons Based on a Generalized Dieterici Equation of State

A new three-parameter Dieterici type equation of state is employed for studying the high-temperature thermodynamic characteristics of hydrocarbons. This generalized equation of state differs from the known Dieterici equation of state by a modified attractive term. That is, a new thermodynamic similarity parameter is introduced in the attractive term of the Dieterici equation of state. The parameters of the equation of state are determined through the experimental values on the critical-point parameters of hydrocarbons. The equation of state is presented in the reduced form, from which follows the single-parameter law of corresponding states. The proposed equation of state gives the value of maximum attainable superheat for hydrocarbons of about 0.887 to 0.894 times the critical temperature. The new three- parameter generalized Dieterici equation of state offers an acceptable compliance with experimental results of maximum attainable superheat of hydrocarbons.

Statistical models of network traffic

The article considers the issues of statistical modeling of traffic in telecommunication networks with packet switching. The simulation results are used in the development of network technical condition management systems, in particular, diagnostics, troubleshooting and network configuration management. The peculiarities of congestion control of separate network segments are emphasized. With improper analysis the overload condition can be mistaken for equipment failure. Therefore, control and elimination of congestion is a statistical task. The concept of end-to-end network diagnostics is considered. This concept provides for effective assessment of the quality of functioning of all network components taking into account their interrelationships. The main issues are the interaction of equipment, inefficient configuration, improper network organization and user operation. Methods of traffic statistical control characteristics based on perforated and marker bucket algorithms are analyzed. A feature of these algorithms is the formation of a strict output stream at a rate that does not depend on the non-uniformity of the input stream. The possibility of improving the token bucket algorithm by adapting to changes in the statistical characteristics of traffic is shown. To solve this problem, statistical mathematical models of network traffic are built. Data traffic circulating in telecommunication networks by packet switching has self-similar (fractal) properties. The self-similar process retains its properties when considered at different time scales (invariance to scale changes). The degree of statistical stability of the process with multiple scaling is determined by the Hirst parameter (the self-similarity parameter). Graphs of statistical characteristics of low-speed and high-speed data traffic are obtained. Their comparative analysis is carried out.

NUMERICAL SIMULATION OF THERMOLUMINESCENCE QUARTZ PARTICLE

One of the widely used methods for studying minerals is the thermoluminescent (TL) method, which is used to date Quaternary sedimentary rocks. Usually, the difficulty in using TL dating is the lack of information about the structure of the mineral used in the experiment. For reliable interpretation of experimental data, the authors applied the digital twin method. In this case, all stages of the transformation of the mineral used are modeled from its burial in sedimentary rocks to the stimulation of the TL-signal splash in it under laboratory conditions on special installations, taking into account the unevenness heat transfer inside the sample of this mineral. The paper presents the results of numerical simulation of the TL signal from a spherical particle of natural quartz. The modeling was carried out in two stages. At the first stage, the influence of the burial time of quartz in the sedimentary rocks on its TL signal was determined. For this, the problem was posed of the accumulation of the crystal lattice quartz, information about the time of its presence in the natural radiation field. At the second stage, the TL signal from a spherical particle was simulated, which was heated in an experimental setup from the surface according to a linear law. It was found that, firstly, the burial time of quartz affects the shape of the TL-curve and, therefore, the mineral is applicable for dating. Secondly, it is necessary to control the heating mode of the quartz sample, since at high heating rates, heat transfer irregularities are significant for particles with a large radius. It also affects the shape of the TL signal and, as a result, can make it difficult to obtain age definitions or distort them. To assess the effect of non-uniformity of heat transfer, a dimensionless similarity parameter was proposed, which connects the radius of the particles and the heating rate.