Measuring, Modeling, and Attacking AI: Toward Efficient and Trustworthy Edge Intelligence

Join us for “Measuring, Modeling, and Attacking AI: Toward Efficient and Trustworthy Edge Intelligence,” Towson AI Institute's Inaugural Research Seminar featuring Anik Mallik, Ph.D., Jess and Mildred Fisher Endowed Professor of Computing Sciences and faculty research lead at the Towson University AI Institute. Click here to register Talk Abstract Artificial intelligence (AI) applications, which once depended on cloud servers, now are in the palm of the masses; credit goes to edge computing and modern edge devices, equipped with higher computing abilities and mobility. However, ensuring less resource (computing and network) utilization and energy consumption, and security and privacy of the edge-AI applications remains a challenge. These research issues need to be addressed for edge intelligence to reach and impact the society on a larger scale. This talk argues that energy is simultaneously the central efficiency bottleneck and an underappreciated security vulnerability in edge AI. To further elaborate this philosophy, first, a measurement and modeling work will be presented in this talk, characterizing energy consumption of AI inference across edge devices with different SoCs, models, and inference delegates. Then an overall performance enhancement strategy will be discussed for edge intelligence, considering wireless networks, device mobility, and battery constraints. In addition, discussion will include how energy and network side channels can allow an adversary to fingerprint the model architecture and application running on an edge device, compromising model privacy with a non-invasive and stealthy attack on edge AI. Finally, the vision of an inclusive society with advanced assistive technologies, enabled by edge intelligence, will be shared with the audience. This research contributes to human lives, as well as to the national economy and security. Learn more about Anik Mallik Anik Mallik is an assistant professor in the Department of Compute