Weekly ML papers

The Case for Learned Index Structures

comment: super wired paper... :(

Gaussian Process Optimization in the Bandit Setting: No Regret and Experimental Design

03/01/2018

comment:

Optimize an unknown, noisy function by formalizing this problem as MAB problem. Solve it through GP optimization. GP-UCB can bound its cumulative regret in terms of maximal information gain.

GP http://cs229.stanford.edu/section/cs229-gaussian_processes.pdf

GP-UCB algorithm