-
tidnabbil
Tidnabbil is a toolkit for addressing multi-armed bandit problems, both in the pure exploration and regret formulations. Such problems include Best Arm Identification, Minimum Threshold Identification, and much more. This toolkit focuses on solving such problems using iterative saddle point methods. (This library supersedes purex_games.)
-
MetaGrad
MetaGrad is an adaptive first-order method for online convex optimisation. By learning the learning rate, MetaGrad significantly outperforms worst-case regret guarantees in many practical examples. It adapts to the structure of the loss functions even without any curvature.
-
Squint
Squint is an advanced method for prediction with expert advice. By learning the learning rate, Squint provides superior regret guarantees in the form of second-order and quantile bounds.
-
SIM-PL
A simulator for digital hardware, used in higher education for courses on digital circuitry and computer architecture. For the University of Amsterdam course Architectuur en Computerorganisatie see here or here. A coursebook (in Dutch) using SIM-PL is available here.