Date of Award
Doctor of Philosophy
Dr. Ala Al-Fuqaha
Dr. Ajay Gupta
Dr. Alvis Fong
Dr. Mohsen Guizani
Smart services are an important element of the Internet of Things (IoT) ecosystem where insights are drawn from raw data through the use of machine learning techniques. However, the pathway to develop IoT smart services is complicated as IoT data presents several challenges for machine learning, including handling big data, shortage of labeled data, and the need to benefit from the spatio-temporal relations hidden in the training data.
In this dissertation, after reviewing the state-of-the-art deep learning (DL) and deep reinforcement learning (DRL) techniques and their use in support of IoT applications, this study proposes to extend DRL to semi-supervised settings using Variational Autoencoders (VAEs). The proposed semi-supervised DRL algorithm exploits the statistical inference of VAEs to benefit from unlabeled data and improve the accuracy of the trained agent to take the best action on the environment.
In IoT applications where most data is spatio-temporal, this study investigates the creation of a dynamic ensemble from distributed deep learning models by considering the spatio-temporal relationships embedded in the training data. The dynamic ensemble does not depend on offline configurations. Instead, it exploits the spatio-temporal relationships embedded in the training data to generate dynamic weights for the underlying weak distributed deep learners to create a stronger learner.
We also present an approach for path planning based on Generative Adversarial Networks (GANs) and crowd-sourced data for wayfinding applications in IoT-enabled environments. In our approach, a GAN is used to recommend accurate and reliable paths to desired destinations. We apply the proposed methods to real-world smart city scenarios including smart energy, smart transportation, and smart buildings. Our results showcase the potential of DL techniques in the development of IoT smart services in general and assert the superiority of our proposed approach compared to baseline methods presented in the recent literature.
Restricted to Campus until
Mohammadi, Mehdi, "Exploring the Role of Semi-Supervised Deep Reinforcement Learning and Ensemble Methods in Support of the Internet of Things" (2018). Dissertations. 3296.