Vår 2021

FYS-3033 Deep learning - 10 stp

The course is administrated by

Institutt for fysikk og teknologi

Type of course

The course is available as a singular course, also to students enrolled at other universities in Norway, exchange students and Fullbright students. The course will only be taught if there are sufficiently many students. Please contact the student adviser as soon as possible if you are interested in following the course.

Course overlap

FYS-8033 Deep Learning 8 stp

Course contents

Deep Learning, a subfield of machine learning, has in recent years achieved state-of-the-art performance for tasks such as image classification, object detection and natural language processing. This course will study recent deep learning methodology such as e.g. convolutional neural networks, autoencoders and recurrent neural networks, will discuss recent advances in the field, and will provide the students with the required background to implement, train and debug these models. There will be a significant practical component, where students will gain hands-on experience. The course will in addition to deep learning algorithms contain elements of image processing, pattern recognition and statistics.

Application deadline

Applicants from Nordic countries: 1 June for the autumn semester and 1 December for the spring semester. Exchange students and Fulbright students: 1 October for the spring semester and 15 April for the autumn semester.

Admission requirements

Admission requirements are a Bachelor's degree in physics or similar education, including specialization in physics worth the equivalent of not less than 80 ECTS credits. Local admission, application code 9371 - singular course at Master's level.

Objective of the course

Knowledge - The student is able to

describe advanced deep learning techniques
describe the development of deep learning
discuss recent developments in the field and develop an understanding for when deep learning might not be the optimal methodology
discuss advanced deep learning techniques for specialized settings

Skills - The student is able to

explain the general idea behind deep learning as well as specific algorithms that are being used
apply the learned material to new applications or problem settings
use deep learning methodology for research and industrial settings using software libraries such as e.g. Theano or TensorFlow
carry out an advanced deep learning project after specifications
make appropriate method and architecture choices for a given application or problem setting

General comtence - The student is able to

give an interpretation of recent developments and provide an intuition of the open questions in the field
give an account of the impact of deep learning in modern society and communicate this to non-experts
implement and apply deep learning methods to applications of her/his choosing

Language of instruction

The language of instruction is English and all of the syllabus material is in English. Examination questions will be given in English, but may be answered either in English or a Scandinavian language.

Teaching methods

Lectures: 30 hours

Exercises: 30 hours

Assessment

Portfolio assessment of project assignments counting about 40 % and a final oral examination counting about 60 %. All modules in the portfolio are assessed as a whole and one combined grade is given.

Assessment scale: Letter grades A-F. F - fail.

Re-sit examination (section 22): There is no access to a re-sit examination in this course.

Postponed examination (sections 17 and 21): Students with valid grounds for absence will be offered a postponed examination. Both postponed laboratory exercises and postponed oral examination are arranged during the semester if possible, otherwise early in the following semester. See indicated sections in Regulations for examinations at the UiT The arctic university of Norway for more information.