The well-known vulnerability of Deep Neural Networks to adversarial samples has led to a rapid cycle of increasingly sophisticated attack algorithms and proposed defenses. While most contemporary defenses have been shown to be vulnerable to carefully configured attacks, methods based on gradient regularization and out-of-distribution detection have attracted much interest recently by demonstrating higher resilience to a broad range of attack algorithms. However, no study has yet investigated the effect of combining these techniques. In this paper, we consider the effect of Jacobian matrix regularization on the detectability of adversarial samples on the CIFAR-10 image benchmark dataset. We find that regularization has a significant effect on detectability, and in some cases can make an undetectable attack on a baseline model detectable. In addition, we give evidence that regularization may mitigate the known weaknesses of detectors to high-confidence adversarial samples. The defenses we consider here are highly generalizable, and we believe they will be useful for further investigations to transfer machine learning robustness to other data domains.
Wireless computer networks are increasing exponentially around the world. They are being implemented in both the unlicensed radio frequency (RF) spectrum (IEEE 802.11a/b/g) and the licensed spectrum (e.g., Firetide [1] and Motorola Canopy [2]). Wireless networks operating in the unlicensed spectrum are by far the most popular wireless computer networks in existence. The open (i.e., proprietary) nature of the IEEE 802.11 protocols and the availability of ''free'' RF spectrum have encouraged many producers of enterprise and common off-the-shelf (COTS) computer networking equipment to jump into the wireless arena. Competition between these companies has driven down the price of 802.11 wireless networking equipment and has improved user experiences with such equipment. The end result has been an increased adoption of the equipment by businesses and consumers, the establishment of the Wi-Fi Alliance [3], and widespread use of the Alliance's ''Wi-Fi'' moniker to describe these networks. Consumers use 802.11 equipment at home to reduce the burden of running wires in existing construction, facilitate the sharing of broadband Internet services with roommates or neighbors, and increase their range of ''connectedness''. Private businesses and government entities (at all levels) are deploying wireless networks to reduce wiring costs, increase employee mobility, enable non-employees to access the Internet, and create an added revenue stream to their existing business models (coffee houses, airports, hotels, etc.). Municipalities (Philadelphia; San Francisco; Grand Haven, MI) are deploying wireless networks so they can bring broadband Internet access to places lacking such access; offer limited-speed broadband access to impoverished communities; offer broadband in places, such as marinas and state parks, that are passed over by traditional broadband providers; and provide themselves with higher quality, more complete network coverage for use by emergency responders and other municipal agencies. In short, these Wi-Fi networks are being deployed everywhere. Much thought has been and is being put into evaluating cost-benefit analyses of wired vs. wireless networks and issues such as how to effectively cover an office building or municipality, how to efficiently manage a large network of wireless access points (APs), and how to save money by replacing an Internet service provider (ISP) with 802.11 technology. In comparison, very little thought and money are being focused on wireless security and monitoring for security purposes.