Beskjeder
The repetition lecture will be Tuesday June 5 10.15-12 in Store Aud. This time has the least number of collisions for you and us. Please check "Timeplan" for the room.
And requests for topics to cover are still welcome.
June 6 we will have a repetition lecture. As we cannot cover all topics, we need your help: which topics do you want us to explain again? Please ask by email to us.
Qualified for the final exam will be all students that have passed mandatory 1 and 2. Please check in Devilry that you have passed these now!
Please give us feedback on the course evaluation form provided by FUI!
There will be no more regular group sessions. If you have questions, send us an email or use canvas. Solutions to the exam from 2017 will be available early June.
Please consider applying to be group teacher for this course in 2019!
In function "getRNNOutputWeights", the b_hy tensor shall have shape [1, vocabulary_size and not shape [1, hidden_state_sizes] as specified in the description. I have uploaded an updated version of oblig3 where this is fixed.
Under undervisningsmateriale you can find the exam from 2017. Solution hints will be uploaded in a couple of weeks.
"Mandatory exercise 3" is out. Note, since it was published late the exercise is NOT mandatory. However, we encourage you to have a look at it. Please feel free to give us feedback on the exercise.
The slides from lecture "recurrent neural network" has been updated. I found an error in the GRU equations. The error was minor and all arguments are still valid.
We are still spending all our time finishing mandatory exercise 3.
In the mean time, here is a sample exam from 2017.
Updated solution hints: here
I have not made an exercise for week 12: "Recurrent neural network". We are therefore cancelling the "Gruppeundervisning". However, I encourage you to have a look at the following videos on YouTube:
https://www.youtube.com/watch?v=DDByc9LyMV8&list=PL9Hr9sNUjfsmEu1ZniY0XpHSzl5uihcXZ&index=27
https://www.youtube.com/watch?v=vI2Y3I-JI2Q&list=PL9Hr9sNUjfsmEu1ZniY0XpHSzl5uihcXZ&index=28
The code can be found here as well:
https://github.com/Hvass-Labs/TensorFlow-Tutorials/
Note: We are still working on mandatory exercise number 3. We will post it as soon as it is ready.
Tollef
The solution has been posted.
Note: A student informed me of an error. When building the convolution layers, you need to edit the code as follows (or download the exercise again).
Change "conv = convLayer2D(data, ...)" to "conv = convLayer2D(conv, ...)"
----------------------------- Updated code ------------------------------
conv = data
for ii in range(len(numbOfFilters)):
layerName = 'convLayer%s' % ii
conv = convLayer2D(conv, ...)
Mandatory assignment 2 is out, and can be downloaded from here. Delivery is to be done in devilry, with deadline set at Friday 23. March.
Good luck.
Due to travel, the weekly exercises related to the convnet-lecture will be available on Wednesday.
The lecture slides from the lecture covering convolution neural networks is posted.
Note: I have made a couple of changes
Exercise for week 5 is out.
To run the jupyter notebook on computers at ifi use (python3): "/opt/ifi/anaconda3/bin/jupyter-notebook"
You can unfortunately not use "jupyterhub.uio.no" as the tensorflow module is not installed.
See instructions under undervisningsmaterie/mandatory_exercises here
It is now possible to start working on Mandatory 1 on Linux machines at IFI.
The exercise set is found here
There are some problems with loading CIFAR-10 data on jupyterhub. We are working on solving them, but in the mean time use /opt/ifi/anaconda2/bin/jupyter-notebook on linux computers at IFI.
If your are not working on Linux, the CIFAR-10 data set can be directly downloaded from https://www.cs.toronto.edu/~kriz/cifar.html
Tuesday 23.1 at 12.15-13 we give a lecture on convolution. The lecture will be in room CAML.
From 13-14 and wednesday 10.15-12 we have ordinary group sessions in Modula.
Link to lecture notes is here
In case many show up on Tuesday, 13.15 bring your laptop if you have one.
If you have registered for the course, but cannot log on to Canvas, send an e-mail to studieinfo@mn.no
Since we have so many student, the first two lectures will be in
Store Aud, KNH
You find the lecture plan, lecture notes, and exercises under Timeplan
Curriculum: selected chapters from deeplearningbook.com and other material provided at lectures.