Beskjeder
If you need clarifications during the examination, you can contact the course instructors using the following e-mail addresses:
For questions 1-4:
Eilif Solberg: eilif.solberg@its.uio.no
For questions 5-8:
Narada Warakagoda: n.d.warakagoda@its.uio.no
Please cc your mail to both.
While e-mail contact is the preferred mode, you can also contact the instructors via telephone if you need immediate response. Telephone numbers can be found in Canvas.
If you have registered for the courses TEK5040 or TEK9040, you will be able to take the trial exam 9. December 0900-1600 Hrs. Just log on to Inspera and take the trial exam. More general information can be found here .
We will publish the trial exam questions and answers on the course web page for the benefit of those who are unable to take the trial exam on the 9th December.
A word of caution: The trial exam lasts 7 hours. However, the proper exam lasts only 4 hours. You will get extra 30 minutes to upload a file in case you want to do so.
The trial exam will be Wednesday the 9th of December, available from 09.00 in Inspera. We will also post a pdf-version later for those not able to participate.
You can find it here when all students have submitted the assignment.
The exam will be a 4.5h digital exam, at the same time as the original exam. We will use the Inspera system, you will log in through your browser. We will have a trial exam so you can get familiar with the format. No collaboration is allowed, but otherwise you may use all available resources.
For those who prefer not taking the exam at home, there are options to book a seat at the university. There are two options here (to my knowledge):
1. Book a place through the university booking system.
2. If you want to sit at ITS, Kjeller, you may send an e-mail to studieinfo@its.uio.no
The university does also have library space where you can't book a spot in advance, but they of course only have a finite capacity. See ...
You can find the past examination problems and solutions here.
You can find the tracking quiz here.
Please enter your email after completing the quiz to get a link to the correct answers.
Solutions to the first two mandatory assignments as well as for some exercise can be found at løsningsforslag
On next Friday (13.11.2020), we will have the last interactive session for this semester. There will be three presentations from PhD students on this day. This is your chance to hear about some additional stuff beyond the curriculum and to show support to your fellow students. The agenda is as follows:
09:15 -09:45 Simen Eide, Stochastic gradient Markov chain Monte Carlo https://arxiv.org/pdf/1907.06986.pdf
09:45 -10:05 Lars Bentsen, Deep RL for Smart Home Energy Management https://arxiv.org/pdf/1909.10165.pdf
10:05-10:15 Break
10:15-10:35 Li Meng V4D: 4D Convolutional Neural Networks for Video-level Representation Learning, https://openreview.net/pdf?id=SJeLopEYDH
1...
You can access the quiz here. After providing the answers enter your email and a link to the answers will be posted to you. Check the correct answers using that link.
Mandatory assignment 3 has been released. You can find it here. The topic of the assignment is Generative Adversarial Imitation Learning (GAIL).
Deadline for submission is 20th November 23:59 Hrs. Any help regarding this assignment, you can contact the instructors through e-mail or Padlet as usual.
An optional exercise on Bayesian deep learning can be found here. I apologize for slightly out of sync release of this exercise.
One of our PhD students will present the paper A closer look at deep policy gradients on Friday. The paper takes a closer look at the methods we used in the second mandatory assignment, and might help get a deeper (or critically-minded) understanding. Please join on zoom at 9.15 if you have the opportunity. Meeting ID: 631 8949 6697. Password is available on Canvas.
Lecture video and slides can be found at Timeplan. There is also a link to a GAN demo script. Try running it to see how the GAN training progress unfolds. It is not optimized in any way, feel free to see if you can improve the convergence by e.g. changing learning learning rates or other hyperparameters, or just see the effect the change has on training.
We were not able to run the quiz session last Friday due to internet failure. Those who are interested can access the quiz from https://www.menti.com/zdn32xbx35. At the end of the quiz, you can submit your email to get the results including the correct answers. Enjoy!
Text sequence processing lecture videos and slides can be accessed from timeplan . There are three videos of which two are fairly long (90 mins). The shorter video covers word embedding, the first longer video covers RNNs and CNNs for sequence processing and the last video covers self-attention/transformers and reinforcement learning. There is an optional exercise also on this topic which is available from here.
I appologize for the interruption today due to an internet failure which lasted for about an hour. We have almost finished the summary and the remaining item was the quiz. I will find a way for you to take the quiz. If you have any question please contact us.
The course curriculum is specified by the lecture slides. That means that you only need to study and understand contents of the lecture slides. Literature specified here are optional. We update the timeplan with these optional literature soon.
The exercise can be downloaded from here. This is an optional exercise, so no submission is required.
Lecture video and slides can be accessed through timeplan.
Note that there are some minor problems of the videos. If you preview them on the server, only sound can be heard. But download to and playing on your PC should work fine.
There are three videos.
- Video 1: From introduction to self-supervised learning (about 1 hour, but players show incorrect length).
- Video 2: From metric learning to meta learning (about 1 hour and 10 mins)
- Video 3: From one shot learning to end (about 12 mins).
Even though the videos are long there are only 40 slides (On average about 3 mins per slide).
The second mandatory assignment can now be downloaded. It is due Thursday October 8th at 23.59 hrs. If possible, use padlet for questions.
Update (23.09.2020, 09.42h): The download now includes the missing requirements.txt file. The requirements are: numpy, tensorflow>=2, matplotlib, gym[box2d]
Update (24.09.2020, 09.30h): I last minute renamed the directory with the source code to 'src'. You need to rename this back to 'car_race' for the imports to work as they should.
Lecture video, slides and notes can now be found in Timeplan. The next mandatory assignment will be on Proximal Policy Optimization, and will be out late tomorrow (Monday).
The lecture video, slide, notes and exercises for this week is now available at Timeplan. I apologize for the delay, hopefully you will appreciate the examples I made! To run the examples yourself, you need a supported PDF-reader. Adobe works, see animate documentation for other supported PDF-readers.