Filters








2,490 Hits in 8.6 sec

Prediction with Expert Advice by Following the Perturbed Leader for General Weights [article]

Marcus Hutter, Jan Poland
2004 arXiv   pre-print
When applying aggregating strategies to Prediction with Expert Advice, the learning rate must be adaptively tuned.  ...  The analysis of the alternative "Follow the Perturbed Leader" (FPL) algorithm from Kalai (2003} (based on Hannan's algorithm) is easier.  ...  Prediction with Expert Advice proceeds as follows. We are asked to perform sequential predictions y t ∈ Y at times t = 1,2,....  ... 
arXiv:cs/0405043v2 fatcat:yudoctlpjzdlhjs6zvapnbnble

Prediction with Expert Advice by Following the Perturbed Leader for General Weights [chapter]

Marcus Hutter, Jan Poland
2004 Lecture Notes in Computer Science  
Keywords Prediction with Expert Advice, Follow the Perturbed Leader, general weights, adaptive learning rate, hierarchy of experts, expected and high probability bounds, general alphabet and loss, online  ...  The theory of Prediction with Expert Advice (PEA) has rapidly developed in the recent past.  ...  Prediction with Expert Advice proceeds as follows. We are asked to perform sequential predictions y t ∈ Y at times t = 1,2,....  ... 
doi:10.1007/978-3-540-30215-5_22 fatcat:4y6k44uze5c4dpvq4kdg4owjry

Move from Perturbed scheme to exponential weighting average [article]

Chunyang Xiao
2015 arXiv   pre-print
Finally, we show with some examples that follow-the-leader style algorithms extend naturally to a large class of structured online problems for which the exponential algorithms are inefficient.  ...  However, recently, another algorithm called follow the perturbed leader is developed and achieved about the same performance.  ...  Conclusion We show that for expert advice problem, the follow perturbed leading expert algorithm could be equivalent to the modern exponential weighted average algorithm by carefully choosing the perturbation  ... 
arXiv:1512.07074v1 fatcat:y2jfeefppbdgter3jkuq23wovi

Adaptive Routing Using Expert Advice

András György, György Ottucsák
2006 Computer journal  
Machine learning algorithms for combining expert advice in sequential decision problems are considered.  ...  The goal of these algorithms is to perform, for any behavior of the system, asymptotically as well as the best expert.  ...  ACKNOWLEDGMENT The authors would like to thank László Györfi for useful discussions.  ... 
doi:10.1093/comjnl/bxh168 fatcat:mnjcaf5abjdsbecsf3stgy75aq

Connections between markets and learning

Yiling Chen, Jennifer Wortman Vaughan
2010 ACM SIGecom Exchanges  
We provide an overview of recent research exploring the striking mathematical connections that exist between market maker mechanisms for prediction markets and no-regret learning.  ...  We describe how these connections can be used in the design of efficient prediction markets over combinatorial outcome spaces.  ...  In general, choosing weights in this manner will prevent all of the algorithm's weight from being placed on a single expert. WM can be written in this form with an entropy term as the regularizer.  ... 
doi:10.1145/1980534.1980540 fatcat:6miyv5x3f5ewda6nuxtwklrgaa

Efficient algorithms for online decision problems

Adam Kalai, Santosh Vempala
2005 Journal of computer and system sciences (Print)  
More importantly, these follow-the-leader style algorithms extend naturally to a large class of structured online problems for which the exponential algorithms are inefficient.  ...  We show that by slightly perturbing the totals and then choosing the best decision, the expected performance is nearly as good as the best decision in hindsight.  ...  Acknowledgments We would like to thank Avrim Blum, Bobby Kleinberg, Danny Sleator, and the anonymous referees for their helpful comments.  ... 
doi:10.1016/j.jcss.2004.10.016 fatcat:rz33fd7qmbc5le6mkslp6l4znq

Efficient Algorithms for Online Decision Problems [chapter]

Adam Kalai, Santosh Vempala
2003 Lecture Notes in Computer Science  
More importantly, these follow-the-leader style algorithms extend naturally to a large class of structured online problems for which the exponential algorithms are inefficient.  ...  We show that by slightly perturbing the totals and then choosing the best decision, the expected performance is nearly as good as the best decision in hindsight.  ...  Acknowledgments We would like to thank Avrim Blum, Bobby Kleinberg, Danny Sleator, and the anonymous referees for their helpful comments.  ... 
doi:10.1007/978-3-540-45167-9_4 fatcat:hwuni2wxgndylfbpggfzkq5vou

Online Learning in Case of Unbounded Losses Using the Follow Perturbed Leader Algorithm [article]

Vladimir V. V'yugin
2010 arXiv   pre-print
In this paper the sequential prediction problem with expert advice is considered for the case where losses of experts suffered at each step cannot be bounded in advance.  ...  We present some modification of Kalai and Vempala algorithm of following the perturbed leader where weights depend on past losses of the experts.  ...  To change from the follow leader framework to DTOL we derandomize our FPL algorithm. 2 Games of prediction with expert advice with unbounded one-step losses We consider a game of prediction with expert  ... 
arXiv:1008.4232v1 fatcat:v3zgiesjtbhmhfjrgo73p3el7i

The Follow Perturbed Leader Algorithm Protected from Unbounded One-Step Losses [chapter]

Vladimir V. V'yugin
2009 Lecture Notes in Computer Science  
In this paper the sequential prediction problem with expert advice is considered for the case when the losses of experts suffered at each step can be unbounded.  ...  We present some modification of Kalai and Vempala algorithm of following the perturbed leader where weights depend on past losses of the experts.  ...  Acknowledgments This research was partially supported by Russian foundation for fundamental research: 09-07-00180-a and 09-01-00709a.  ... 
doi:10.1007/978-3-642-04414-4_8 fatcat:em6pfoeol5cg7k6pqgbb7wkek4

Learning Volatility of Discrete Time Series Using Prediction with Expert Advice [chapter]

Vladimir V. V'yugin
2009 Lecture Notes in Computer Science  
In this paper the method of prediction with expert advice is applied for learning volatility of discrete time series.  ...  For merging different expert strategies in a strategy of the learner, we use some modification of Kalai and Vempala algorithm of following the perturbed leader where weights depend on current gains of  ...  Acknowledgments This research was partially supported by Russian foundation for fundamental research: 09-07-00180-a and 09-01-00709-a.  ... 
doi:10.1007/978-3-642-04944-6_3 fatcat:3fcrmeyzgvelfphbpmms3pvnfm

Adaptive Online Prediction by Following the Perturbed Leader [article]

Marcus Hutter, Jan Poland
2005 arXiv   pre-print
When applying aggregating strategies to Prediction with Expert Advice, the learning rate must be adaptively tuned.  ...  The analysis of the alternative "Follow the Perturbed Leader" (FPL) algorithm from Kalai & Vempala (2003) (based on Hannan's algorithm) is easier.  ...  The FPL (Follow the Perturbed Leader) approach solves the second problem by adding to each expert's loss s i <t a random perturbation.  ... 
arXiv:cs/0504078v1 fatcat:vwyj63gmkrecvptbqgwy4cvs3y

Random-Walk Perturbations for Online Combinatorial Optimization

Luc Devroye, Gabor Lugosi, Gergely Neu
2015 IEEE Transactions on Information Theory  
RANDOM-WALK PERTURBATIONS FOR PREDICTION WITH EXPERT ADVICE In this section we refine our method to obtain bounds of optimal order in the special case of prediction with expert advice, that is, when the  ...  Algorithm 2 Random-walk perturbations for prediction with expert advice. Initialization: set L i,0 = 0 and Z i,0 = 0 for all i = 1, 2, . . . , N .  ...  Thus, we are left with the problem of bounding Similarly to [20] , we do this by introducing for all u ∈ S and studying the relationship between the distributions p t andp t .  ... 
doi:10.1109/tit.2015.2428253 fatcat:p4rcrr6i4beohmqmccjteff4lm

Extracting certainty from uncertainty: regret bounded by variation in costs

Elad Hazan, Satyen Kale
2010 Machine Learning  
Prediction from expert advice is a fundamental problem in machine learning.  ...  In this paper we resolve this question, and prove such bounds in the fully adversarial setting, in two important online learning scenarios: prediction from expert advice, and online linear optimization  ...  Acknowledgements We thank Martin Zinkevich for initial discussions on the possibility of variation bounds on the regret.  ... 
doi:10.1007/s10994-010-5175-x fatcat:ebjmfvg6ovfd3bnnfxdnrzsbfm

Follow the Leader with Dropout Perturbations

Tim van Erven, Wojciech Kotlowski
2014 Annual Conference Computational Learning Theory  
We consider online prediction with expert advice.  ...  The two most popular algorithms are Hedge/Weighted Majority and Follow the Perturbed Leader (FPL).  ...  Acknowledgments Tim van Erven was supported by NWO Rubicon grant 680-50-1112, Wojciech Kotłowski by the Foundation for Polish Science under the Homing Plus Program, and Manfred K.  ... 
dblp:conf/colt/ErvenK14 fatcat:dsbthrgonzb2zlopm7iidx6ure

Following the Perturbed Leader for Online Structured Learning

Alon Cohen, Tamir Hazan
2015 International Conference on Machine Learning  
We investigate a new Follow the Perturbed Leader (FTPL) algorithm for online structured prediction problems.  ...  To better understand FTPL algorithms for online structured learning, we present a lower bound on the regret for a large and natural class of FTPL algorithms that use logconcave perturbations.  ...  There, the authors have shown a reduction from the problem of shortest path to prediction with expert advice.  ... 
dblp:conf/icml/CohenH15 fatcat:23gsvqzpmzenbmhmivf6us3yx4
« Previous Showing results 1 — 15 out of 2,490 results