Artificial Intelligence for the Internet of Moving Things

This is a sponsored post from Stevens Institute of Technology.

Written by R. Chandramouli and K.P Subbalakshmi
Department of Electrical and Computer Engineering & Stevens Institute for Artificial Intelligence
Stevens Institute of Technology

Internet of Moving Things (IoMT) denotes connected cars, trucks, trains, etc., and, in general, wireless Internet of connected things that move. Medium to high speed mobility presents several connectivity challenges not experienced by the Internet of (stationary) Things (IoT). This includes rapid variations in the wireless link quality, high user density (e.g., onboard WiFi), significant wireless bandwidth required for video streaming, and wireless coverage holes. A heterogeneous network (HetNet) architecture consisting of multiple radio access technologies (multi-RAT) such as 3G/4G cellular, satellite, and track-side or road-side WiFi is a promising approach for the IoMT.

Artificial Intelligence (AI) algorithms learn to optimize decisions under uncertainty from observed data.  AI tools can be implemented by a software defined IoMT network controller to improve the resilience, capacity, security and other key metrics of an IoMT system.

Cognitive Mobile Networking

A connected car can communicate with the Internet cloud or other cars over one of several wireless links such as 4G LTE, DSRC (dedicated short range communications), satellite or WiFi. A multi-RAT modem in the car periodically measures several parameters (e.g., bit error rate) related to all these links, summarizes the measurement statistics and transmits it to a software defined network (SDN) controller. AI algorithm(s) implemented by the SDN controller ingests the summary statistics to predict the best wireless access link for the next few time slots. This prediction is then used by the modem to switch to the corresponding wireless network. The AI algorithm must be robust against missing measurement statistics, erroneous measurement and other impairments.

Cognitive Mobile Computing

Consider the real-time sensor data based computations required by a fully autonomous self-driving car. These computations have several constraints such as deadlines, dependencies between multiple software or application subcomponents, CPU requirement, memory requirement, etc. An AI algorithm can the (a) recognize the computations that must be run locally in the car and those that need to be offloaded to the cloud, (b) what fraction of the available wireless access links must be used for computational offloading, and (c) learn continually by profiling the applications, networks and the key performance metrics.

These are the types of topics that students explore in the Stevens Online MBA Program. Engineering leaders need a solid understanding of the potential of technology, data and analytics, with an emphasis on teaching students how to mine large amounts of data and apply these insights. Learn more about the artificial intelligence and its related programs at Stevens.

Resources

S. Eman Mahmoodi and K.P. Subbalakshmi, “A time-adaptive heuristic for cognitive cloud offloading in multi-RAT enabled wireless devices,” IEEE Transactions on Cognitive Communications and Networking, vol. 2, no. 2, pp. 194-207, June 2016.

S. Eman Mahmoodi, K.P. Subbalakshmi, and R.N. Uma, Spectrum-awareness in mobile computing – Convergence of cloud computing and cognitive networking, Springer, 2018 (forthcoming).