lunduniversity.lu.se

Automatic Control

Faculty of Engineering, LTH

Denna sida på svenska This page in English

Real-Time and Embedded Systems with Application to Machine Learning

FRT160F

Instructor: Prof. Zonghua Gu, Zhejiang University, Hangzhou, China, http://www.cs.zju.edu.cn/people/zgu/ 

Number of credits: 5 hp

During March 2017, Prof. Zonghua Gu from Zhejiang University will visit Lund University and give a PhD course on "Real-Time and Embedded Systems with Application to Machine Learning". The course is suitable both for students with an interest in real-time and embedded (incl cloud computing and dynamic resource management) and students with an interest in machine learning and deep neural networks.

The course consists of six half-day (4 hour) lectures with the following content:

  • Day 1: Introduction to real-time and embedded systems; basic concepts of real-time operating systems and real-time scheduling; fixed-priority and Earliest Deadline First (EDF) scheduling; Resource synchronization protocols: PIP, PCP and SRP. 
  • Day 2: Servers for handling aperiodic workloads: Polling, Deferrable and Sporadic Servers for fixed-priority scheduling, and Total Bandwidth/Constant Bandwidth Servers for EDF scheduling. 
  • Day 3: Limited Preemptive Scheduling; elastic task model for handling overloads; multicore scheduling (partitioned or global). 
  • Day 4: Introduction to machine learning and artificial intelligence, especially deep neural networks (DNN). Techniques for implementing DNN in resource-constrained embedded systems: pruning, reduced precision, computational offloading, etc. 
  • Day 5: Introduction to HW/SW codesign and HW/SW partitioning; HW/SW implementations of DNNs: GPU, DSP, FPGA and ASIC. 
  • Day 6: Neuromorphic computing with Spiking Neural Networks; IBM TrueNorth as representative example; spiking vs. non-spiking NN. 


The lectures are scheduled in the seminar room at 13.15–17.00 in the afternoons of March 6, 8, 13, 15, 20, 22. The course will go into some technical details and proofs but will not be math-heavy and should be accessible to most students. Students that have taken, e.g., FRTN01 Real-Time Systems at the Department of Automatic Control will recognize parts of the first lecture but nothing else, hence this course will be a nice complement.

Literature:

  1. Textbook: Giorgio Buttazzo, Hard Real-Time Computing Systems, Springer, 2011 
  2. A reading list of research papers will be provided in class. 


Student performance is assessed in the form of a module examination. During the written exam (90 min), students need to answer questions based on presented lectures to demonstrate their understanding of intelligent embedded systems.

Page Manager: