Search for: All records

On-device ML is increasingly used in different applications. It brings convenience to offline tasks and avoids sending user-private data through the network. On-device ML models are valuable and may suffer from model extraction attacks from different categories. Existing studies lack a deep understanding of on-device ML model security, which creates a gap between research and practice. This paper provides a systematization approach to classify existing model extraction attacks and defenses based on different threat models. We evaluated well known research projects from existing work with real-world ML models, and discussed their reproducibility, computation complexity, and power consumption. We identified the challenges for research projects in wide adoption in practice. We also provided directions for future research in ML model extraction security. 

A promising avenue for improving the effectiveness of behavioral-based malware detectors is to leverage two-phase detection mechanisms. Existing problem in two-phase detection is that after the first phase produces borderline decision, suspicious behaviors are not well contained before the second phase completes. This paper improves CHAMELEON, a framework to realize the uncertain environment. CHAMELEON offers two environments: standard–for software identified as benign by the first phase, and uncertain–for software received borderline classification from the first phase. The uncertain environment adds obstacles to software execution through random perturbations applied probabilistically. We introduce a dynamic perturbation threshold that can target malware disproportionately more than benign software. We analyzed the effects of the uncertain environment by manually studying 113 software and 100 malware, and found that 92% malware and 10% benign software disrupted during execution. The results were then corroborated by an extended dataset (5,679 Linux malware samples) on a newer system. Finally, a careful inspection of the benign software crashes revealed some software bugs, highlighting CHAMELEON's potential as a practical complementary antimalware solution. 

By 2018, it is no secret to the global networking community: Internet of Things (IoT) devices, usually controlled by IoT applications and applets, have dominated human lives. It has been shown that popular applet platforms (including If This Then That (IFTTT)) are susceptible to attacks that try to exfiltrate private photos, leak user location, etc. As new attacks might show up very frequently, tracking them fast and in an efficient and scalable manner is a daunting task due to the limited (e.g., memory, energy) resources at the IoT/mobile device and the large network size. Towards that direction, in this paper we propose a decentralized Dynamic Information Flow Tracking (DDIFT) framework that overcomes these challenges, better adapts to the IoT context, and further, is able to illuminate IoT applet attacks. In doing so, we leverage the synergy between: (i) a dynamic information flow tracking module that considers the application of tags with different types along with provenance information and runs in the mobile device at a fast timescale, (ii) a forensics analysis module running in the cloud at a slow timescale, (iii) distributed optimization to optimize various functionalities of the above modules as well as their interaction. We show that our framework is able to detect IoT applet attacks with higher accuracy (on average 81% improvement for different URL upload attack scenarios) and decreases resource wastage (on average 71% less memory usage under different integrity attack scenarios) compared to traditional DIFT, opening new horizons for IoT privacy and security.