Advanced Edge Computing Architectures for IoT

Recommended Prerequisites

Computer Architecture, Computer programming, Machine Learning.

Teaching Methods

The teaching methods of the curricular unit are organized around two complementary components: theoretical or seminar (S) and practical or tutorial (OT). Theoretical seminar classes are mainly intended to expose content by the teacher and to clarify doubts of general interest to the class. The study and analysis of works describing the state-of-the-art and their discussion in an expository way in a classroom environment are also foreseen. The tutorial classes (OT) intend to consolidate the concepts presented in the seminar classes (S) through the development of a practical project.

Learning Outcomes

The objective of this course is to introduce theoretical and practical concepts of edge computing, presenting its advantages when compared to centralized computing. With the widespread adoption of Internet of Things (IoT) systems, the amount of data constantly being generated has increased significantly. This curricular unit intends to explore the distributed processing of such amounts of data at the edge to ensure a more efficient use of the available computational resources, such as a more efficient use of memory, bandwidth, decreased latency and energy consumption.

The processing of massive amounts of IoT data often requires the use of distributed machine learning techniques, such as federated learning, which provides increased levels of data security and privacy. In particular, the course will emphasize the acquisition of computing skills that involve the use of low-power GPUs and System-on-Chip (SoC) at the edge.

Work Placement(s)

No

Syllabus

- Evolution of distributed systems: Why computing at the edge?

- Multi-tier distributed system architectures (examples in IoT and 5G networks)

- State-of-the-art hardware and software edge computing platforms

- Edge GPU / System-on-Chip computing:

     => Memory management

     => Workload balancing between host and device

     => Energy consumption / management

- Taking machine learning out of datacenters:

     => Federated learning

     => Reinforcement learning on the edge

- Advanced edge computing applications:

     => Edge-supported smart cities

     => Healthcare

     => Assistive robotics

     => Manufacturing and industry 4.0

     => Edge-supported smart vehicles and drones

     => Augmented and virtual reality devices

- Energy and resource management, workload balancing and distribution in multi-tier fog computing systems.

Head Lecturer(s)

Paulo José Monteiro Peixoto

Assessment Methods

Assessment
Mini Tests: 20.0%
Report of a seminar or field trip: 30.0%
Research work: 50.0%

Bibliography

1. Thomas Sterling, Matthew Anderson, and Maciej Brodowicz. High Performance Computing: Modern Systems and Practices, 1st Edition, Morgan Kaufmann, 5th December 2017.

2. Kirk, David B., and W. Hwu Wen-Mei. Programming massively parallel processors: a hands-on approach. Morgan kaufmann, 2016.

3. Buyya, Rajkumar, and Satish Narayana Srirama, eds. Fog and edge computing: principles and paradigms. John Wiley & Sons, 2019.

4. Qiang Yang, Yang Liu, Yong Cheng, Yan Kang, Tianjian Chen, and Han Yu. Federated Learning, Morgan & Claypool, 2019.

5. Pete Warden, and Daniel Situnayake. TinyML. O'Reilly Media, Inc., December 2019. ISBN: 9781492052043.