Learning Rotations Learning rotations with little regret
نویسندگان
چکیده
We describe online algorithms for learning a rotation from pairs of unit vectors in R. We show that the expected regret of our online algorithm compared to the best fixed rotation chosen offline over T iterations is O( √ nT ). We also give a lower bound that proves that this expected regret bound is optimal within a constant factor. This resolves an open problem posed in COLT 2008. Our online algorithm for choosing a rotation matrix is essentially an incremental gradient descent algorithm over the unbounded set of all matrices, with specially tailored projections. We also show that any deterministic algorithm for learning rotations has Ω(T ) regret in the worst case.
منابع مشابه
Corrigendum to “ Learning rotations with little regret ” September 7 , 2010
There is an unfortunate error in our paper “Learning rotations with little regret” [HKW10] which appeared in COLT 2010. The sampling procedure for the noise matrix given in [HKW10] does not produce matrices with the right density. In this corrigendum, we describe the error, and give a correct sampling procedure. Unfortunately, even with the correct sampling procedure, the regret bound we get is...
متن کاملMinimax Algorithm for Learning Rotations
It is unknown what is the most suitable regularization for rotation matrices and how to maintain uncertainty over rotations in an online setting. We propose to address these questions by studying the minimax algorithm for rotations and begin by working out the 2-dimensional case. The problem of online learning of rotations is defined as follows. In every iteration t = 1, 2, . . . , T , the lear...
متن کاملLearning Rotations
Many different matrix classes have been tackled recently using online learning techniques, but at least one major class has been left out: rotations. We pose the online learning of rotations as an open problem and discuss the importance of this problem.
متن کاملAcquiring new spatial intuitions: learning to reason about rotations.
There are certain simple rotations of objects that most people cannot reason about accurately. Reliable gaps in the understanding of a fundamental physical domain raise the question of how learning to reason in that domain might proceed. Using virtual reality techniques, this project investigated the nature of learning to reason across the domain of simple rotations. Learning consisted of the a...
متن کاملVisual-spatial ability, learning modality and surgical knot tying.
INTRODUCTION The ability to mentally rotate an object in 3 dimensions has been shown with an individual's score on the Vandenberg and Kuse Mental Rotations Test. The was to determine whether this Mental Rotations Test could be used to predict performance complex surgical skill - the tying of a 1-handed surgical reef knot. In addition, we learning a spatially complex surgical skill could be achi...
متن کامل