Date of Award

8-13-2019

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

First Advisor

Dr. Ashwin Ashok

Second Advisor

Dr. Raj Sunderraman

Third Advisor

Dr. Zhipeng Cai

Fourth Advisor

Dr. Raheem A. Beyah

Abstract

The proliferation of Internet-of-Things (IoT) devices has opened a plethora of opportunities for smart networking, connected applications and data driven intelligence. The large distribution of IoT devices within a finite geographical area and the pervasiveness of wireless networking present an opportunity for such devices to collaborate. Centralized decision systems have so far dominated the field, but they are starting to lose relevance in the wake of heterogeneity of the device pool. This thesis is driven by three key hypothesis: (i) In solving complex problems, it is possible to harness unused compute capabilities of the device pool instead of always relying on centralized infrastructures; (ii) When possible, collaborating with neighbors to identify security threats scales well in large environments; (iii) Given the abundance of data from a large pool of devices with possible privacy constraints, collaborative learning drives scalable intelligence.

This dissertation defines three frameworks for these hypotheses; collaborative computing, collaborative security and collaborative privacy intelligence. The first framework, Opportunistic collaboration among IoT devices for workload execution, profiles applications and matches resource grants to requests using blockchain to put excess capacity at the edge to good use. The evaluation results show app execution latency comparable to the centralized edge and an outstanding resource utilization at the edge. The second framework, Integrity Threat Identification for Distributed IoT, uses a new spatio-temporal algorithm, based on Local Outlier Factor (LOF) uniquely using mean and variance collaboratively across spatial and temporal dimensions to identify potential threats. Evaluation results on real world underground sensor dataset (Thoreau) show good accuracy and efficiency. The third frame- work, Collaborative Privacy Intelligence, aims to understand privacy invasion by reverse engineering a user’s privacy model using sensors data, and score the level of intrusion for various dimensions of privacy. By having sensors track activities, and learning rule books from the collective insights, we are able to predict ones privacy attributes and states, with reasonable accuracy. As the Edge gains more prominence with computation moving closer to the data source, the above frameworks will drive key solutions and research in areas of Edge federation and collaboration.

DOI

https://doi.org/10.57709/15048551

File Upload Confirmation

1

Download

Share

COinS