Beskjeder
The exam results are on their way out today.
A solution hint which is used when evaluating the exams will soon be published here.
For those who passed the exam, asking for a justification("begrunnelse") of the grade will not give insight to your full score, only refer to the solution hint and tell if you are close to the boundary of pass/fail. This is done as instructed by the faculty.
During the correction, is you suggested other valid arguments compared to those mentioned in the solution hint, you were also given full score.
Monday at 14.15 we have scheduled a zoom if you have questions:
Join Zoom Meeting
https://uio.zoom.us/j/67302421200
As we get a large number of questions, and also repeated questions already answered, we have made a document where all questions asked are included in the exercise text.
From now on, this document will be updated when new questions are answered.
A notification here will be made when the document is updated.
UPDATED tuesday 22.00
Link to document with all questions
2c: Could you explain in a different way what you mean?
Exercise text: Discuss whether a CNN is efficient for detecting long range dependencies in an image.
A: Long range dependencies in an image: dependencies between pixels with a large spatial distance within the image.
3a: Do you mean RNN, not CNN?
Exercise text: How could you structure a CNN to be used for 1D time series data with a
stream of incoming data?
A: CNN is correct. Not the formulation "stream of incoming data".
1e: Do you mean weight initialization or during training: we mean initilaization. (As long as you state you interpretation of the question both interpretations will get full score.)
What are the weights for each exercise?
As the front page states, each exercise has the same weight. When scoring, we give each exercise a score from 0 to 10, which we use to compute an average score over the exercise set.
1c: Can you comment if the updated weight is smaller or larger? We do not want to give any hints, but an error in the sign will only result in a small reduction of the score for that exercise.
3a: when working on a 1D stream of sequential data, are you talking about that not all the training data is available at the same time?
Consequences of using a stream of data will stream of sequential data will be in terms of what kind of architecture you can use for the problem due to the...
Question: Did you mean RNN instead of CNN?
Answer: No, we mean CNN.
Question: What do you mean by "long range dependencies in an image"?
Answer: Think of long range dependencies as distant information e.g. opposite sides of an image.
Q: Are we talking about sin() activation function as general or do we need to discuss if it is good for task 1b?
We meant in general.
Given a question like this, you can state your interpretation of the question in your answer, and you will get full score for a good answer to your stated interpretation.
Is is also fine to ask us.
There is a small notation error in W[2]
Correct notation is
/studier/emner/matnat/ifi/IN5400/v20/beskjeder/oppg2b.png
Rules state that you can only contact us through MN Studieinfo with your questions, and only questions related to clarification/interpretation/errors will be answered.
Contact MN studieinfo at
https://www.mn.uio.no/om/hms/koronavirus/brukerstotte-eksamen-v20.html
Only those type of questions will be answered during the zoom session too.
June 3 at 14.15 we will be available in case you have questions.
We will only answer questions from you, if you have some.
The duration will depend on the number of questions.
Here is the zoom link:
Join Zoom Meeting
https://uio.zoom.us/j/66967631040
The exam will contain exercises similar to previous year, but be larger to correspond to two working days done in a 7-day period.
The exam exercise setwill consist of a PDF-file in Inspera.
You will get requests to prepare ONE file (PDF or zip) to upload. The exam text will tell you the type of file you should submit.
The exam must be your individual work.
PhD-students: threshold for pass will be B as usual. Please aim to answer all questions as perfect as possible to make sure you pass.
To prepare for the exam, solve previous year's exams. The exercises and solution hints are now under material/examexercises.
Group sessions will now be on request as 1-to-1 zoom sessions with Milena, Anne or Tollef. Send an email to request.
The repetition lecture will be 27.5 on zoom.
Please email us your wishes about topics to explain again, or previous exam exercises to explain.
The next lecture will be given as an interactive lecture on Zoom April 29. The topic will be reinforcement learning.
https://uio.zoom.us/j/9869400981
The guest lecture on General Adversarial Networks will be cancelled due to the guest lecturer being assigned to work on the COVID app that Simula works on. GANs will be taken out from the curriculum.
The next lecture on unsupervised learning will be given as an interactive lecture on Zoom April 22, but all material and external links and exercises will be uploaded this week to enable you to prepare questions. You will find the material on /studier/emner/matnat/ifi/IN5400/v20/material/Unsupervised%20learning/index.html
We have also set up one-to-one meetings with us to get help with Mandatory 2. Sign up in the doodle given in a previous email.
The exam will be a home exam with grades pass/fail.
You will have 7 days available from June 3 to do work corresponding to a workload of 1-2 days.
The first week of material for digital teaching is now available.
Presently, we record the lectures from home, and try digital interactive sessions to discuss them.
We are currently modifying Mandatory 2 to avoid using GPUs as due to the crises we can not use the solution we had planned for.
You will need Pytorch and Anaconda on your laptop. If for some reason you cannot install it and do not have computer access with python/DL-software, inform us by email and we will give an alternative exercise.
As both you and we get more experience will digital teaching, please share good suggestions from e.g. other courses with us either on Piazza or by email. We definitely learn from week to week, but please help us improve!
We will share information about exams as soon as we get it. What is clear now is that no exams will take part in Silurveien, but the alternatives are not clear.
Starting from thursday 12.3, all lectures will only be made available on video due to new requirements from UiO due to the COVID-19 situation. Groups will also be cancelled the coming weeks, but we will find new forums for questions. In the mean time, use Piazza to ask any questions.
Please follow the web pages for new information.
It is under material/mandatory/mandatory_1, see link here
We plan to publish Mandatory 1 Monday 10.2 before noon.
You will have three weeks to complete it.
Update 29.11: 5 spots were opened at 12.00 today.
Monday 27.1 at 12.00 5 more students can register for the course. At 12.00, these go to the first 5 who register.
The first lecture is Wednesday January 15 at 10.15, and the first group Monday January 20 at 12.15