Artificial neural networks are models inspired by the mammal brain. Several artificial neurons in a network communicating similarly to the brain, make out a powerful model that has been shown to be able to approximate virtually any continuous function. During the 1990s, the interest in artificial neural networks declined, due to limited processing power and data availability, and other technologies often worked better. In the 2000s, the availability of big data and the advent of faster computers (and graphics processing units able to perform fast mathematical computations) has made it possible to train significantly larger models. These models gave rise to the term Deep Learning, and has outperformed the previous state of the art in tasks such as image recognition and natural language processing.
Rough course outline
- Recurrent Neural Networks
- Convolutional Neural Networks
- Unsupervised and semi-supervised methods (Auto-Encoders, Ladder Networks)
- Regularization (Norm penalty, Dropout, Batch normalization)
- Paper presentations
The course will be carried out as a flipped classroom, which means that every participant has the mandatory homework of watching video lectures at home before attending the course sessions. At the sessions, we will have material to help discussing the content.
This is a PhD course targeting students who are interested in machine learning and artificial neural networks. Some mathematical maturity is expected, and a basic course in linear algebra and machine learning (equivalend to TDA 231 Algorithms for Machine Learning & Inference
or FFR135 Artificial Neural Networks
) are required before taking this course.
The number of seats in this course is limited to 30 students. Master students will be accepted on a first come, first served basis.
If you as a master student want to take the course for credits, we ask you to clear this with the student office, to get yourself registered with a course code
. For students that complete the course, we will be able to certify this, provided that the necessary administration has been taken care of. We will not be able to solve your problem if you do not have a course code.
You should be registered on one of the following course code, Chalmers: DAT235
, GU: DIT575
The examination is split in two parts. For passing the course, you have to attend at least 80% of the discussion sessions. For 3(G), point 1 must be fulfilled, and a presentation with reasonable clarity will have to be performed in the end of the course. For excellent presentations, that shows a deeper understanding and excellent presentation skills, the grade 4 or 5(VG) can be obtained.
- A written summary will have to be submitted before each discussion session, summarizing the content that you have watched on video lectures at home. The summary should be between 400 and 500 words. These submissions should also contain a brief comment on what you considered most difficult.
- In the end of the course, each participant will have to present a relevant research paper. We will provide a list of suggestions.
Please join our group: firstname.lastname@example.org
for announcements and discussions.