rbardenet / bml-course Goto Github PK

Shouldn't it be - n * theta[2] - ... instead of - theta[2] - ...?
(where \theta[2] is log(σ), and n the sample size)

I understood during the course that Bayesian optimization is better suited than other classical optimization (e.g gradient-descent) when the objective function is "hard" to calculate.

I don't really understand this concept of "hard", does it mean we do not know a precise analytic of the function (black-box) ? Computationally hard ?
For example, if I know the analytic form of an objective function with random variables, could it be considered "hard" as the function is not deterministic ? Would Bayesian optimization be suited in this case ?

Example question: raise here any question you want to discuss publicly.

It is possible that some students may discover typos or notation problems when reading the different Lecture Notes.

To alleviate this problem, what do you think about grouping the Latex "In progress" in this document ?

This is just a proposal but it might allow us to optimize the quality of the Lecture notes.

Hi,

We have been wondering about the next class: the MVA planning says there's a class next week (Friday, March 13th). What is it supposed to be about? Some other students have been wondering if this is supposed to be for oral presentations, but it seems no further details were given on that subject.

Thank you