Content deleted Content added
convert keypoints from talk page Special:Diff/947318998 into prose text. A low sample complexity is desired in robotics which can be realized with metric learning. |
|||
Line 87:
==Other Settings==
In addition to the supervised learning setting, sample complexity is relevant to [[semi-supervised learning]] problems including [[active learning]],<ref name="Balcan">{{cite journal |doi = 10.1007/s10994-010-5174-y|title = The true sample complexity of active learning|journal = Machine Learning|date = 2010|volume = 80|issue = 2–3|pages = 111–139|first = Maria-Florina|last = Balcan}}</ref> where the algorithm can ask for labels to specifically chosen inputs in order to reduce the cost of obtaining many labels. The concept of sample complexity also shows up in [[reinforcement learning]],<ref>{{citation |last = Kakade | first = Sham | title = On the Sample Complexity of Reinforcement Learning | place = University College London | publisher = Gatsby Computational Neuroscience Unit. | series = PhD Thesis | year = 2003 | url = http://www.ias.tu-darmstadt.de/uploads/Research/NIPS2006/SK.pdf}}</ref> [[online machine learning|online learning]], and unsupervised algorithms, e.g. for [[dictionary learning]].<ref>{{cite journal |last1 = Vainsencher | first1 = Daniel | last2 = Mannor | first2 = Shie | last3 = Bruckstein | first3 = Alfred | title = The Sample Complexity of Dictionary Learning | journal = Journal of Machine Learning Research | volume = 12 | pages = 3259–3281 | date = 2011 | url = http://www.jmlr.org/papers/volume12/vainsencher11a/vainsencher11a.pdf}}</ref>
==Efficiency in robotics==
A high sample complexity means, that many calculations are needed for running a [[Monte Carlo tree search]].<ref>{{cite conference |title=Monte-carlo tree search by best arm identification |author=Kaufmann, Emilie and Koolen, Wouter M |conference=Advances in Neural Information Processing Systems |pages=4897--4906 |year=2017 }}</ref> Its equal to a [[Model-free (reinforcement learning)|model free]] brute force search in the state space. In contrast, a high efficiency algorithm has a low sample complexity.<ref>{{cite conference |title=The chin pinch: A case study in skill learning on a legged robot |author=Fidelman, Peggy and Stone, Peter |conference=Robot Soccer World Cup |pages=59--71 |year=2006 |publisher=Springer }}</ref> Possible techniques for reducing the sample complexity are [[metric learning]]<ref>{{cite conference |title=Sample complexity of learning mahalanobis distance metrics |author=Verma, Nakul and Branson, Kristin |conference=Advances in neural information processing systems |pages=2584--2592 |year=2015 }}</ref> and model based reinforcement learning.<ref>{{cite journal |title=Model-ensemble trust-region policy optimization |author=Kurutach, Thanard and Clavera, Ignasi and Duan, Yan and Tamar, Aviv and Abbeel, Pieter |journal=arXiv preprint arXiv:1802.10592 |year=2018 }}</ref>
==References==
| |||