CAREER: Wireless InferNets: Enabling Collaborative Machine Learning Inference on the Network Path

Machine learning is increasingly being integrated into wireless network applications, such as video surveillance, smart healthcare and industrial internet of things, to derive actionable intelligence from the rich data collected or generated by wireless devices. To support real-time machine learning and decision making in these applications, a large amount of data has to be transferred from the data source to the destination and processed by an inference algorithm in a timely manner. Traditionally, data transfer and machine learning inference are treated as two separate optimization tasks, but such approaches are inefficient as they ignored the interaction between data transfer and machine learning inference, resulting in either a large data transfer delay or a large inference delay. This CAREER project overcomes the limitations of existing approaches by proposing wireless InferNets, a new wireless network architecture that enables collaborative machine learning inference among network nodes on the data transfer path. The successful completion of this CAREER project will promote the understanding of the synergy between distributed inference and networking, and catalyze a paradigm shift of future wireless networks to support emerging applications and services in security, healthcare and other technological domains. The project also contains a significant educational component and provides abundant opportunities to nurture and attract students, especially from underrepresented groups, to engage in computer science and engineering. This CAREER project develops models, algorithms and protocols to realize the core functions of wireless InferNets and address challenges caused by network and device heterogeneity, dynamic and imperfect network states, and multiple user contention for the limited resources via three main research aims. Specifically, it aims at (i) developing practical distributed inference routing algorithms and protocols and conducting theoretical analysis to understand the performance limits; (ii) developing new multi-armed bandit algorithms to perform inference routing with uncertain network information; (iii) developing distributed multi-agent deep reinforcement learning algorithms for inference routing in multi-user wireless InferNets. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.