Tuesday Morning Tutorial - Questions

Disclaimer: Dieser Thread wurde aus dem alten Forum importiert. Daher werden eventuell nicht alle Formatierungen richtig angezeigt. Der ursprüngliche Thread beginnt im zweiten Post dieses Threads.

Tuesday Morning Tutorial - Questions
Last week I attended the tuesday morning tutorial (14.7.20) instead of my usual tutorial because Katjas tutorial did not take place. Some questions arised during the tutorial which Max couldn’t answer at that time. I was wondering if the answers to the questions have been posted somewhere?

Unfortunately I do not remember the questions but I remember being interested in the answers :smiley:

I answered them in the next tutorial (21.7.20). Here you go:

1. In exam SS2019, question 4.2, are formulas needed or is a qualitative description sufficient:

Answer: The ideal solution would consist of a sketch of the example annotated with the relevant variables, the formulae into which these variables enter and a brief explanation of the formulae and the overall algorithm. Also, don’t forget that part of the points can already be gained by explaining when and why the backpropagation algorithm is needed. That being said, formulae are not strictly necessary if the explanation is clear. However, the most time-efficient way to answer this question will probably involve stating the formulae instead of producing a wall of text.

2. What exactly constitutes the basic rules of Bayesian inference. In assignment 2.1 there are only 4 (Bayes’ rule is not among them) whereas in the reference solution to the exam there are six.

At the time of assignment 2.1 Bayes’ rule had not been covered yet in the lecture - hence the ommission from the solution. The exam solution is the reference solution. Please note: the rules stated here are also basic rules of probability theory in general, not just of Bayesian reasoning.

Regarding 4.2, I would also like to point out that “approximate formulas” can be a pretty concise way of describing things, and save you from remembering the fiddly details. For example, writing down the update formula but instead of also writing down the formula for [m]\Delta[/m] saying something like “where [m]\Delta[/m] is a term representing the error/loss”.