Environment of Multi-Armed Bandit

For more info about our services contact : help@bestpfe.com

Table of contents

I General Introduction and State of the art 
1 Introduction 
1.1 Motivation
1.2 Modeling
1.3 Thesis Outline
2 Multi-Armed Bandit 9
2.1 Environment of Multi-Armed Bandit
2.1.1 Stationary Bandit
2.1.2 Adversary Bandit
2.1.3 Contextual Bandit
2.1.4 Linear Bandit
2.2 Gittins index
2.3 The strategy of trade-off
2.3.1 Thompson Sampling
2.3.2 Boltzmann Exploration
2.3.3 Upper Confidence Bound
2.3.4 Epsilon-Greedy
2.4 Regret Lower bound
2.5 Pure Exploration and Best Armed Identification
3 Bandit with side information 
3.1 Multi-class Classification with Bandit feedback
3.1.1 Multiclass Classification
3.1.2 Algorithms for Multi-class Classification with Bandit Feedback
3.2 Multi-label Classification with Bandit feedback
3.2.1 Multilabel Classification
3.2.2 Algorithm for Multi-label Classification with Bandit feedback
4 Multi-Objective Multi-Armed Bandit 
4.1 Multi-Objective Optimization
4.1.1 Front Pareto setting
4.1.2 Dominance method
4.1.3 Aggregation method
4.2 Multi-Objective Optimization in Bandit environment
4.2.1 Algorithms for MOMAB
II Contributions 
5 Passive-Aggressive Classification with Bandit Feedback 
5.1 Multi-class PA with Bandit feedback
5.1.1 Simple PAB
5.1.2 Full PAB
5.1.3 Experiments
5.1.4 Conclusion
5.2 Bandit feedback in Passive-Aggressive bound
5.2.1 Analysis
5.2.2 Experiments
5.2.3 Conclusion
5.3 Bandit feedback with kernel
5.3.1 BPA Online Kernel
5.3.2 Kernel Stochastic Gradient Descent with BPA loss
5.3.3 Experiments
5.4 Bandit PA algorithm for Multi-label Classification
5.4.1 Preliminaries
5.4.2 Analysis
5.4.3 Experiments
6 Optimized Identification algorithm for ²-Pareto Front 
6.1 Identification the ²-Pareto front
6.2 Experiments
IIIConclusion and Perspectives
7 Conclusions 
7.1 Summary of contributions
7.2 Research in the future