Block Switching: A Stochastic Approach for Deep Learning Security

Abstract

Recent study of adversarial attacks has revealed the vulnerability of modern deep learning models. That is, subtly crafted perturbations of the input can make a trained network with high accuracy and produce arbitrary incorrect predictions, while maintaining imperceptible to human vision system. In this paper, we introduce Block Switching (BS), a defense strategy against adversarial attacks based on stochasticity. BS replaces a block of model layers with multiple parallel channels, and the active channel is randomly assigned in the run time, hence unpredictable to the adversary. We show empirically that BS leads to a more dispersed input gradient distribution and superior defense effectiveness compared with other stochastic defenses such as stochastic activation pruning. Compared to other defenses, BS is also characterized by the following features: (i) BS causes less test accuracy drop; (ii) BS is attack-independent; and (iii)

Monitoring the Spatio-Temporal Distribution of Soil Salinity Using Google Earth Engine for Detecting the Saline Areas Susceptible to Salt Storm Occurrence

“…On the other hand, the gamma parameter determines the 'spread' or reach of the kernel. A lower gamma value means that the influence of a single training sample extends farther, while a higher gamma value results in a more localized influence [54,55].…”

Critical Failure Mode Determination of Steel Moment Frames by Plastic Analysis Optimization Principles

“…To deal with this problem, artificial intelligence (AI) can be an appropriate choice. Generally, machine learning and deep learning approaches are two branches of AI, where numerous studies in different areas exist in the literature [19,20]. Artificial intelligence (AI) and machine learning (ML) are being increasingly used in structural engineering applications for tasks such as structural analysis, design optimization, and predictive maintenance [21][22][23][24].…”

“…Kaveh et al [20,29] and Kohama et al [30] have presented studies on the determination of the failure load factor of plane frames employing the genetic algorithm. Kaveh and Jahanshahi have studied the limit plastic procedure of frames by means of the ant colony system and other similar algorithms [31][32][33].…”