Learning Observable Operator Models via the Efficient Sharpening Algorithm
Herbert Jaeger, Mingjie Zhao, Klaus Kretzschmar, Tobias Oberstein, Dan Popovici, Andreas Kolling · The MIT Press eBooks · 2006
This chapter contains sections titled: 14.1 Introduction, 14.2 The Basic Ideas behind Observable Operator Models, 14.3 From HMMs to OOMs: Matrix Representations of OOMs, 14.4 OOMs as Generators and Predictors, 14.5 Understanding Matrix OOMs by Mapping Them to Functional OOMs, 14.6 Characterizing OOMs via Convex Cones, 14.7 Interpretable OOMs, 14.8 The Basic Learning Algorithm, 14.9 History, Related Work, and Ramifications, 14.10 Overview of the Efficiency Sharpening Algorithm, 14.11 The Efficiency Sharpening Principle: Main Idea and a Poor Man's ES Learning Algorithm, 14.12 Essentials of Suffix Trees, 14.13 A Suffix-Tree-Based Version of Efficiency Sharpening, 14.14 A Case Study: Modeling One Million Pounds, 14.15 Discussion, Appendix A: Proof of Proposition 14.4, Property 4, Appendix B: Proof of Proposition 14.5, Appendix C: The Three OOMs from Figure 14.2, Appendix D: Proof of Proposition 14.10, Appendix E: Proof of Proposition 14.12, Appendix F: Proof of Proposition 14.13, Appendix G: Proof of Proposition 14.14, Appendix H: Proof of Proposition 14.15, Appendix I: Finding Good Characteristic Events, Appendix J: Running Invalid OOMs as Sequence Generators, Appendix K: Details of the One Million Pound Learning Experiment, Notes