Adversarial example detection for DNN models: a review and experimental comparison

Deep learning (DL) has shown great success in many human-related tasks, which has led to its adoption in many computer vision based applications, such as security surveillance systems, autonomous vehicles and healthcare. Such safety-critical applications have to draw their path to success deployment once they have the capability to overcome safety-critical challenges. Among these challenges are the defense against or/and the detection of the adversarial examples (AEs). Adversaries can carefully craft small, often imperceptible, noise called perturbations to be added to the clean image to generate the AE. The aim of AE is to fool the DL model which makes it a potential risk for DL applications. Many test-time evasion attacks and countermeasures, i.e., defense or detection methods, are proposed in the literature. Moreover, few reviews and surveys were published and theoretically showed the taxonomy of the threats and the countermeasure methods with little focus in AE detection methods. In this paper, we focus on image classification task and attempt to provide a survey for detection methods of test-time evasion attacks on neural network classifiers. A detailed discussion for such methods is provided with experimental results for eight state-of-the-art detectors under different scenarios on four datasets. We also provide potential challenges and future perspectives for this research direction.

     

Study of Magnetic Properties of Fe-Doped CuO: Monte Carlo Simulations

Using Monte Carlo simulations, we are studying the magnetic properties of Fe-doped CuO thin films. The total magnetizations and the susceptibilities are studied as a function of the effect doping, external magnetic field, and exchange coupling. The critical temperature is discussed as a function of the effect of iron concentration. On the other hand, we investigate the effect of increasing temperatures on the coercive field for a constant value of exchange coupling and a fixed concentration. The coercive magnetic field is found to decrease with increasing temperature values until reaching its null value. The effect of increasing the exchange coupling amount on the saturation magnetic field H _s is illustrated. A linear growth of the saturation magnetic field is found as a function of the exchange coupling interaction. To complete this study, we presented and discussed the magnetic hysteresis cycle loops.     

On dual-phase-lag thermoelasticity theory with memory-dependent derivative

A new mathematical model of generalized thermoelasticity with memory-dependent derivatives for the dual-phase-lag heat conduction law is constructed. The governing equations of the new model are applied to a half-space subjected to ramp-type heating. Laplace transforms technique is used. The solution is obtained for different types of functions representing the thermal shock and for different values of the parameter of the time fraction derivative of the model. The effects of time-delay and arbitrary kernel function on elastic material are studied and represented graphically. The predictions of the theory are discussed and compared with dynamic classical coupled theory.     

The Use of Acoustic Emission Intensity Analysis for the Assessment of Cover Crack Growth in Corroded Concrete Structures

In this paper, acoustic emission (AE) intensity analysis was utilized to assess the concrete cover cracking due to steel corrosion in reinforced concrete structures. A total of 30 reinforced concrete prism samples were tested under an accelerated corrosion test coupled with continuous AE monitoring using attached AE sensors. The samples were cast with three concrete cover thicknesses (20, 30 and 40 mm) around steel bars and were exposed to five percentages of steel mass loss: 1, 2, 3, 4, and 5 %. The cover cracking was monitored daily by visual inspection to detect and measure crack widths. Different AE signal parameters were continuously recorded during the tests, including number of hits, signal strength, energy, and amplitude. The acquired AE events were subjected to an intensity analysis of signal strength to estimate historic index (H (t)) and severity (\(S_r)\). In addition, a b value analysis was conducted on all AE data and the results were compared to those obtained from the intensity analysis. The results showed that increasing the cover thickness had no significant impact on AE parameters (number of hits, cumulative signal strength, cumulative energy, amplitude, H (t), and \(S_{r})\) at similar values of crack growth. Nonetheless, varying the cover thickness from 20 to 40 mm resulted in lower crack widths and slightly higher b values at the same levels of steel mass loss. It was also found that both H (t) and \(S_r\) showed a more evident correlation with the values of crack growth than did b values, regardless of cover thickness or percentage of steel mass loss. Finally, an intensity classification chart was developed to quantify the cover crack growth based on the values of H (t) and \(S_{r}\).     

Effect of initial composition on phase selection in Ni–Si powder blends processed by mechanical alloying

The effect of initial powder blend composition on the synthesis and formation mechanism of nickel silicide phases was investigated by mechanical alloying in Ni-60 and Ni-66.7?at.% Si powder blends. It was noted that the equilibrium NiSi phase started to form in the early stages of milling and that the amount of the NiSi phase in the milled powder increased with increasing milling time. Even though, under equilibrium conditions, a mixture of both the NiSi and NiSi₂ phases was expected to be present in the Ni-60?at.% Si composition and the stoichiometric NiSi₂ phase in the Ni-66.7?at.% Si composition, the NiSi phase was present in both the compositions investigated. However, while only the NiSi phase was present homogeneously in the Ni-60?at.% Si powder blend, both the NiSi phase and a very small amount of unreacted Si were present in the powder blend of Ni-66.7?at.% Si composition. This unexpected phase constitution in the milled powders was attributed to a partial loss of Si during mechanical alloying of the powder blends, confirmed by energy dispersive X-ray spectrometer analyses, and explained on a thermodynamic basis.     

Assessing abrasion performance of self-consolidating concrete containing synthetic fibers using acoustic emission analysis

The key objective of this investigation was to evaluate the abrasion resistance of self-consolidating concrete (SCC) with and without synthetic fibers (SynFs). The abrasion resistance of normal concrete was also investigated in this study for comparison. The abrasion test was performed on concrete specimens according to the rotating-cutter method along with continuous monitoring of acoustic emission (AE) using attached AE sensors. The effects of changing concrete type and incorporating various types (flexible and semi-rigid) and lengths of SynFs on the abrasion behaviour were investigated with the aid of AE analysis. AE signal characteristics such as amplitude, signal strength, number of hits, and duration were gathered during testing. Furthermore, the collected AE data was used to complete b-value analysis as well as intensity analysis resulting in three additional parameters: b-value, severity (S_r), and historic index (H(t)). The results showed that the AE parameters were directly correlated with the abrasion damage in all tested mixtures. Adding SynFs to all SCC mixtures enhanced their abrasion resistance. The flexible fibers variety exhibited better abrasion performance on average than the semi-rigid fibers. Meanwhile, longer fibers showed lower abrasion resistance than the shorter ones with the same type. The results also indicated that AE intensity analysis was able to determine the ranges for H(t) and S_r that identify the extent of damage due to abrasion of SynF-reinforced SCC.