Grant: Discovery Grants Program - Individual

The rapid advancement of mobile/edge computing has led to the widespread use of mobile devices. Recently, a novel sensing paradigm, known as Mobile Crowdsensing (MCS), had emerged. In MCS, a group of mobile users is recruited by the platform to perform real-time and large-scale tasks, such as traffic monitoring, healthcare services, and environment assessment. Meanwhile, Deep Learning (DL) have made remarkable breakthroughs in various domains. There is a growing demand for on-device DL services within the MCS framework. To enhance mobile users' privacy and optimize computational efficiency, an Edge-Cloud Collaborative (ECC) approach has been devised for MCS. In ECC-based MCS system, a large DL model is divided into two components. The shallow portion is deployed on mobile devices as a feature extractor, while the larger portion resides in the cloud for result inference. However, this system faces two significant challenges. First, there is a potential risks of mobile users' privacy breaches during task allocation and real-time inference. Second, users' dynamic and heterogeneous nature makes effective task allocation a particularly complex issue. To position Canadian researchers as global leaders in edge-cloud computing and practical AI applications in the big data era, there is a critical need to explore efficient and privacy-preserving techniques for task allocation and model partition in ECC-based MCS services. The main objective of this proposed research is to tackle the aforementioned challenges by exploring a range of advanced privacy-preserving and interdisciplinary methods, including DL, cryptographic algorithms, and other relevant approaches. More specifically, this proposal will concentrate on the following issues in ECC-based MCS system: 1.Thoroughly examine and comprehend potential threat models and various attacking techniques. 2.Develop effective privacy protection technologies for model partitioning and data transmission. 3.Propose solutions for task allocation that are both privacy-preserving and efficient. 4.Investigate and compare evaluation metrics to systematically assess the trade-off between privacy and utility. 5.Introduce novel evaluation measurements by combing DL metrics and cryptographic assumptions. 6.Evaluate the practicality and viability of the proposed framework in diverse real-world MCS scenarios, such as industrial 4.0, e-Healthcare, and intelligent transportation systems. This proposed research will heavily leverage the extensive research expertise of the Principal Investigator, reinforced by substantial support from the Master of Business Analytics and Maser of Computing & Data Analytics programs at Saint Mary's University. The anticipated outcomes of this research can create fundamental and innovative concepts that advance privacy-enhanced techniques and AI-driven mobile applications. Additionally, it can contribute significantly to the training and education of Highly Qualified Personnel in Canada.