Messages

For questions about the exam e-mail me or Anne-Marie. The admin also have my phone number, I think.

use meet.google.com/kpd-vdhp-bus

The vide lecture for the November 17th session is available here.

The practical session on November 17 will be pre-recorded. There will be in-person lecturing on November 24th however. 

https://uio.zoom.us/j/6566829641

From and including Wednesday, Oct. 13th, there will be in-person teaching of the practical sessions on Wednesdays in the seminar room "Prolog", 12:00-14:00 (new time due to clashes with other activities for most students).

You are reminded that the project has a final deadline on October 22.

You were supposed to submit a preliminary report by the end of last week. If you have not yet done so, please submit one this week. This will be used to guide you through the project (it is not graded individually). So please make sure to submit it to get feedback now, before it is too late to get any feedback.

Christos

The practical session on October 5th will be held by Dirk in seminar room Java in Ole-Johan Dahls hus in-person.

Dear all,

Dirk is back from holidays, so he is going to do programming-focused sessions on Tuesdays. I will do theory-focused sessions on Wednesdays instead. There will be no more mix of theory and programming from now on in my sessions, unless it is to illustrate something.

C.

Correct src/privacy/randomised_response.py so that it takes into account the number of features and randomises all of them, so as to release a randomised version of the full data set in a way that satisfies epsilon-differential privacy.

In particular, you must take care to properly take into account the fact that each feature release counts as an additional query.

Next week we will go into the Laplace and Exponential mechanisms, and there will be a practice session with Dirk.

Chapter 3 from the notes.

Chapters 1-3 from "The Algorithmic Foundations of Differential Privacy"

I have added two tutorial videos on using TFP for inference, focusing on:

I. Specifying graphical models either by

1. manually specifying a joint distribution
2.  using the JointDistributionXXX TFP classes

II. Performing MAP inference

III. Performing MCMC inference.

I do not go into the theory, but the basic idea of MAP and MCMC inference is explained, as well as how you specify what to condition on. 

The notebooks are here:

https://github.com/olethrosdc/ml-society-science/blob/master/src/bayesian-inference/TFP-Tutorial.ipynb

https://github.com/olethrosdc/ml-society-science/blob/master/src/bayesian-inference/JointNamedExample.ipynb

TFP is a bit finicky, so sometimes you get weird errors even if you try and something simple.

You should all join a Project group of 3-5 people on Canvas:

https://uio.instructure.com/courses/33326/groups#tab-6798

You should all submit a preliminary report outlining what you are going to do in your project by Friday, October 1. The earlier you submit something concrete there, the better chances you have of successfully completing your project.

Sections 2.4-2.6.

Please join a group for assignment 2. We will keep working on this next week.

https://uio.instructure.com/courses/33326/groups#tab-6797

Tomorrow we will go over these notebooks:

https://github.com/olethrosdc/ml-society-science/blob/master/src/reproducibility/Pipeline.ipynb

https://github.com/olethrosdc/ml-society-science/blob/master/src/reproducibility/Regression.ipynb

https://github.com/olethrosdc/ml-society-science/blob/master/src/reproducibility/KNN.ipynb

Please complete the first assignment on Canvas. It is not graded, but it is used as a background knowledge check.

https://mattermost.uio.no/ifi-undervisning/channels/instk5000

Notes are here:

https://github.com/olethrosdc/ml-society-science/blob/master/notes.pdf

Chapter 1, 2 of the notes.

The two first videos by Dirk.

All zoom sessions will be live. Dirk's pre-recorded lecturers cover python basics necessary for the project.

The link for Dirk¡¯s pre-recorded sessions doesn¡¯t seem to work. Kindly use this one instead:
https://www.youtube.com/playlist?list=PL_TKn86MxCqbejw4c3LB-I4Oqv3WDS5Py

Here is the canvas page for the course:

https://uio.instructure.com/courses/33326

Here is the zoom link: 

https://uio.zoom.us/j/69466413655

Dirk will be on paternity leave during August and September, so his first lessons were pre-recorded and are available here. The materials are here, and the exercises here. Dirk will teach in person at UiO from October on.