Syllabus for |
|
| FFR135 - Artificial neural networks |
| Artificiella neurala nätverk |
| |
| Syllabus adopted 2020-02-20 by Head of Programme (or corresponding) |
| Owner: MPCAS |
|
|
7,5 Credits
|
| Grading: TH - Pass with distinction (5), Pass with credit (4), Pass (3), Fail |
| Education cycle: Second-cycle |
|
Major subject: Bioengineering, Chemical Engineering, Engineering Physics
|
|
Department: 16 - PHYSICS
|
Teaching language: English
Application code: 11113
Open for exchange students: Yes
Block schedule:
B+
Maximum participants: 200
| Module |
|
Credit distribution |
|
Examination dates |
|
Sp1 |
Sp2 |
Sp3 |
Sp4 |
Summer course |
No Sp |
| 0100 |
Examination |
7,5 c |
Grading: TH |
|
7,5 c
|
|
|
|
|
|
|
26 Oct 2020 am L 
|
05 Jan 2021 pm J, |
19 Aug 2021 pm J |
In programs
MPDSC DATA SCIENCE AND AI, MSC PROGR, Year 1 (elective)
MPDSC DATA SCIENCE AND AI, MSC PROGR, Year 2 (elective)
MPSYS SYSTEMS, CONTROL AND MECHATRONICS, MSC PROGR, Year 2 (elective)
MPENM ENGINEERING MATHEMATICS AND COMPUTATIONAL SCIENCE, MSC PROGR, Year 2 (elective)
MPENM ENGINEERING MATHEMATICS AND COMPUTATIONAL SCIENCE, MSC PROGR, Year 1 (compulsory elective)
MPCAS COMPLEX ADAPTIVE SYSTEMS, MSC PROGR, Year 1 (compulsory)
Examiner:
Bernhard Mehlig
Go to Course Homepage
Eligibility
General entry requirements for Master's level (second cycle)
Applicants enrolled in a programme at Chalmers where the course is included in the study programme are exempted from fulfilling the requirements above.
Specific entry requirements
English 6 (or by other approved means with the equivalent proficiency level)
Applicants enrolled in a programme at Chalmers where the course is included in the study programme are exempted from fulfilling the requirements above.
Course specific prerequisites
Analysis in one variable, linear algebra, basic skills in analysis in several variables, and programming.
Aim
This course describes how neural networks are used in machine learning. Neural networks are distributed computational models inspired by the structure of the human brain, consisting of many simple processing elements that are connected in a network. Neural networks have revolutionised how we solve important problems in the engineering sciences, such as image analysis (object recognition and location), prediction, and control. The course gives an overview and a basic understanding of currently used neural-network algorithms, and exhibits similarities as well as differences between these methods. The main emphasis of this introductory course is on three connected topics: recurrent (Hopfield) networks, supervised learning with deep neural networks, and unsupervised learning (reinforcement learning). The goal is to explain how and why the algorithms work, when and how they fail, how to program the standard methods from scratch, and how to use packages that allow to easily set up and to efficiently run larger networks.
Learning outcomes (after completion of the course the student should be able to)
-distinguish between supervised and unsupervised learning, explain the key principles of the corresponding algorithms, understand differences and similarities
-understand under which circumstances neural-net algorithms are the method of choice
-understand and explain strengths and weaknesses of the neural-net algorithms
-implement the algorithms introduced in class on a computer, both from scratch and using neural-net packages
-interpret the results of computer simulations and communicate conclusions in a clear, logical, and concise fashion
-understand the historical development of the field of machine learning with neural networks
-reflect ethical questions posed by machine learning, as well as possible risks
Content
The course is based on Machine learning with neural networks https://arxiv.org/abs/1901.05639.
1. Statistical mechanics of neural nets
McCulloch-Pitts neurons, Hopfield nets, stochastic optimisation, Boltzmann machines
2. Deep learning
Perceptrons, backpropagation, stochastic gradient descent, deep learning, recurrent nets
3. Unsupervised learning
Hebbian learning, radial basis-function nets, reinforcement learning
Organisation
Lectures
Homework problems
Programming with programming language of choice (commonly matlab or python). We use OpenTA for the homework problems.
Exercise classes
Homework problems and exam questions
Guest lectures
From research and/or industry, possibilities for MSc theses
Short instruction videos
Literature
Course book
B. Mehlig Machine learning with neural networks https://arxiv.org/abs/1901.05639
Additional references
I. Goodfellow, Y. Bengio & A. Courville, Deep Learning https://www.deeplearningbook.org
J. Hertz, A. Krogh & R. G. Palmer, Introduction to the theory of neural computation, Addison-Wesely, Redwood City (1991).
S. Haykin, Neural Networks: a comprehensive foundation, 2nd ed., Prentice Hall, New Jersey (1999)
R.S. Sutton & A. Barto, Reinforcement learning: An Introduction, 2nd ed., MIT Press, http://www.incompleteideas.net/book/the-book-2nd.html
Examination including compulsory elements
The final grade is based on homework assignments (50%) as well as on a written examination (50%). To pass the course one must obtain more than a given minimum number of points in the written exam.