This course aims to study algorithms for settings that lie between the fully stochastic and fully adversarial setting. The first half of this course will review classical algorithms for the adversarial setting (e.g., EXP3/4, mirror descent, etc.) and the stochastic setting (UCB, action elimination, experimental design etc.) to get everyone on the same page and to develop a toolkit. In the second half we will focus on settings that lie between these two extremes with the aim of practical, sample efficient, and robust algorithms. In particular, we will highlight algorithms that obtain the "best of both worlds" that adapt to the unknown setting. We will consider settings where the parametric model is correct, but the user's preferences are allowed to drift or switch over time. We will also consider settings when the model is merely approximately correct, as well as settings where an adversary can arbitrarily corrupt some small number of observations. Finally, we will consider censored and delayed observations.
The Best of Both Worlds: Part II movie free download hd
Download
https://rempwebbuylin.blogspot.com/?wr=2wHc8w
eebf2c3492