END-TO-END OPTIMIZATION WITH EDGE INTELLIGENCE FOR RELIABLE WIRELESS IOT COMMUNICATIONS
Abstract
In this proposal, Deep Neural Networks (DNNs) will become the primary tool due to the capability to understand the underlying system topology using adequate learning representations. Hyper-parameters are an integral part of the DNN module and consist of: learning rate, number of iterations, number of hidden layers, number of neurons, choice of activation; sigmoid, rectified linear unit (RELU) or tanh functions. For proper functionality of DNN, the hyper-parameters must be carefully selected. Selecting DNN hyperparameters is not trivial, and hence, a model is needed. Water quality wireless sensor communication can be considered as an end-to-end reconstruction problem, in which the payload transmitted over a wireless channel is reconstructed at the receiver. In wireless communication, this is known as autoencoder, which represents the entire wireless communication system and jointly optimizes the transmitter and receiver over the wireless channel. The goal of this project is to achieve an end-to-end global optimization of reliable wireless IoT communications in a challenging tropical rural lake environment by interpreting the considered communications system as an auto-encoder model and using belief propagation deep learning (DL)-DNN Bayesian black-box optimization with Gaussian process (BBBO-GP). Moreover, in order to reduce the computational complexity at the IoT devices side, as well as reducing network traffic and latency, the edge computing technique will be used in this project.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 11, 2021
- Source ID
- FA23862014045
Entities
People
- Rosdiadee Nordin
Organizations
- Air Force Office of Scientific Research
- National University of Malaysia
- United States Air Force