Complete Defense Framework to Protect Deep Neural Networks against Adversarial Examples

Although Deep Neural Networks (DNNs) have achieved great success on various applications, investigations have increasingly shown DNNs to be highly vulnerable when adversarial examples are used as input. Here, we present a comprehensive defense framework to protect DNNs against adversarial examples. First, we present statistical and minor alteration detectors to filter out adversarial examples contaminated by noticeable and unnoticeable perturbations, respectively. Then, we ensemble the detectors, a deep Residual Generative Network (ResGN), and an adversarially trained targeted network, to construct a complete defense framework. In this framework, the ResGN is our previously proposed network which is used to remove adversarial perturbations, and the adversarially trained targeted network is a network that is learned through adversarial training. Specifically, once the detectors determine an input example to be adversarial, it is cleaned by ResGN and then classified by the adversarially trained targeted network; otherwise, it is directly classified by this network. We empirically evaluate the proposed complete defense on ImageNet dataset. The results confirm the robustness against current representative attacking methods including fast gradient sign method, randomized fast gradient sign method, basic iterative method, universal adversarial perturbations, DeepFool method, and Carlini & Wagner method.

Safety climate and injury occurrence of repair, maintenance, minor alteration and addition works

Purpose – The repair, maintenance, minor alteration and addition (RMAA) sector has been expanding in many developed cities. Safety problems of the RMAA sector have attracted the attention of many governments. This study has the objectives of comparing the level of safety climate of workers, supervisors and managers in the RMAA sector; and explaining/predicting the impact of safety climate on injury occurrence of workers, supervisors and managers. Design/methodology/approach – A questionnaire survey was administered to RMAA contracting companies in Hong Kong. Findings – When comparing the safety climate perception of workers, supervisors and managers in the RMAA sector, the supervisors group had the lowest mean safety climate score. Results showed that a positive workforce safety attitude and acceptance of safety rules and regulations reduced the workers' likelihood of having injuries. A reasonable production schedule led to a lower probability of supervisors being injured. Management commitment and effective safety management reduced the probability of managers being injured. Originality/value – This study revealed variations of safety climate at the different levels in the organizational hierarchy and their varying influence on safety performance of the RMAA sector. Safety of RMAA works could be improved by promulgating specific safety measures at the different hierarchy levels.