%0 Book %1 LaSze19:book %A Lattimore, T. %A Szepesvári, Cs. %D 2020 %I Cambridge University Press (to be printed soon) %K bandits %T Bandit Algorithms %U https://tor-lattimore.com/downloads/book/book.pdf %X Multi-armed bandits have now been studied for nearly a century. While research in the beginning was quite meandering, there is now a large community publishing hundreds of articles every year. Bandit algorithms are also finding their way into practical applications in industry, especially in on-line platforms where data is readily available and automation is the only way to scale. We had hoped to write a comprehensive book, but the literature is now so vast that many topics have been excluded. In the end we settled on the more modest goal of equipping our readers with enough expertise to explore the specialized literature by themselves, and to adapt existing algorithms to their applications. This latter point is important. Problems in theory are all alike; every application is different. A practitioner seeking to apply a bandit algorithm needs to understand which assumptions in the theory are important and how to modify the algorithm when the assumptions change. We hope this book can provide that understanding. What is covered in the book is covered in some depth. The focus is on the mathematical analysis of algorithms for bandit problems, but this is not a traditional mathematics book, where lemmas are followed by proofs, theorems and more lemmas. We worked hard to include guiding principles for designing algorithms and intuition for their analysis. Many algorithms are accompanied by empirical demonstrations that further aid intuition. We expect our readers to be familiar with basic analysis and calculus and some linear algebra. The book uses the notation of measure-theoretic probability theory, but does not rely on any deep results. A dedicated chapter is included to introduce the notation and provide intuitions for the basic results we need. This chapter is unusual for an introduction to measure theory in that it emphasizes the reasons to use σ-algebras beyond the standard technical justifications. We hope this will convince the reader that measure theory is an important and intuitive tool. Some chapters use techniques from information theory and convex analysis and we devote a short chapter to each.

@book{LaSze19:book, abstract = {Multi-armed bandits have now been studied for nearly a century. While research in the beginning was quite meandering, there is now a large community publishing hundreds of articles every year. Bandit algorithms are also finding their way into practical applications in industry, especially in on-line platforms where data is readily available and automation is the only way to scale. We had hoped to write a comprehensive book, but the literature is now so vast that many topics have been excluded. In the end we settled on the more modest goal of equipping our readers with enough expertise to explore the specialized literature by themselves, and to adapt existing algorithms to their applications. This latter point is important. Problems in theory are all alike; every application is different. A practitioner seeking to apply a bandit algorithm needs to understand which assumptions in the theory are important and how to modify the algorithm when the assumptions change. We hope this book can provide that understanding. What is covered in the book is covered in some depth. The focus is on the mathematical analysis of algorithms for bandit problems, but this is not a traditional mathematics book, where lemmas are followed by proofs, theorems and more lemmas. We worked hard to include guiding principles for designing algorithms and intuition for their analysis. Many algorithms are accompanied by empirical demonstrations that further aid intuition. We expect our readers to be familiar with basic analysis and calculus and some linear algebra. The book uses the notation of measure-theoretic probability theory, but does not rely on any deep results. A dedicated chapter is included to introduce the notation and provide intuitions for the basic results we need. This chapter is unusual for an introduction to measure theory in that it emphasizes the reasons to use σ-algebras beyond the standard technical justifications. We hope this will convince the reader that measure theory is an important and intuitive tool. Some chapters use techniques from information theory and convex analysis and we devote a short chapter to each.}, added-at = {2020-03-17T03:03:01.000+0100}, author = {Lattimore, T. and {Sz}epesv{\'a}ri, {Cs}.}, bdsk-url-1 = {http://proceedings.mlr.press/v54/hanawal17a.html}, biburl = {https://www.bibsonomy.org/bibtex/202fed2c828a0a14217377c7843bb53a5/csaba}, date-added = {2019-07-20 14:35:41 -0600}, date-modified = {2020-03-07 15:04:59 -0700}, interhash = {211c1b71adfe8dadacd4906f5140c223}, intrahash = {02fed2c828a0a14217377c7843bb53a5}, keywords = {bandits}, month = {August}, publisher = {Cambridge University Press (to be printed soon)}, timestamp = {2020-03-17T03:03:01.000+0100}, title = {Bandit Algorithms}, url = {https://tor-lattimore.com/downloads/book/book.pdf}, year = 2020 }