Algorithms for Learning and Search in Structured Prediction
Kursbeschreibung
| Studiengang | Modulkürzel | Leistungs- bewertung |
|---|---|---|
| BA-2010 | AS-CL | 8 LP |
| NBA | AS-CL | 8 LP |
| Master | SS-CL, SS-TAC | 8 LP |
| Magister | - | - |
| Dozenten/-innen | Artem Sokolov |
| Veranstaltungsart | Hauptseminar |
| Erster Termin | 21.04.2015, 11:15 |
| Zeit und Ort | Di, 11:15–12:45, INF 327 / SR 5 (SR) |
| Commitment-Frist | 16.06.–15.07.2015 |
Leistungsnachweis
Regular and active participation, oral presentation.
Sprache
English
Inhalt
As a rule in Structured Prediction and, in particular, in NLP one views inference, decoding and learning as independent subroutines. Given the approximate nature of all of them and imprecise modeling assumptions, this may lead to poor performance. Moreover, because of fundamental complexity results and because of limited resources, in majority of important tasks achieving exactness for these fundamental processes is hopeless. In this seminar we will study structured prediction, reinforcement learning and combinatorial optimization approaches that view search space exploration and learning in concert, analyse cases when exactness of inference and learning cannot be guaranteed and that try to optimally balance speed and precision.
Kursübersicht
Seminarplan
| Datum | Materialien | Referent |
| 21.04 | Introduction (slides) | Artem Sokolov |
| 28.04 | Hal Daume III, Daniel Marcu: Learning as search optimization: approximate large margin methods for structured prediction, ICML'05 | reading group (Artem Sokolov) |
| 05.05 | Thomas Finley, Thorsten Joachims: Training Structural SVMs when Exact Inference is Intractable. ICML'08 Alex Kulesza, Fernando Periera: Structured Learning with Approximate Inference. NIPS'07 | reading group (Juri Opitz) |
| 12.05 | Hal Daume III, John Langford: Search-based Structured Prediction. Machine Learning Journal'09 | Julia Kreutzer |
| 19.05 | Ross, Gordon, Bagnell: A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning AISTATS'11 | reading group (Artem Sokolov) |
| 26.05 | Jiarong Jiang, Adam Teichert, Hal Daumé III, Jason Eisner: Learned Prioritization for Trading Off Accuracy and Speed NIPS'12 | reading group (Artem Sokolov) |
| 02.06 | He He, Hal Daume III, Jason Eisner: Imitation Learning by Coaching. NIPS'12 He He, Hal Daume III, Jason Eisner: Dynamic Feature Selection for Dependency Parsing. EMNLP'13 | reading group (Mayumi Ohta) |
| 09.06 | M. Chang and L. Ratinov and D. Roth: Structured Learning with Constrained Conditional Models. Machine Learning Journal'12 | Juri Opitz |
| 16.06 | Ofer Meshi, David Sontag, Tommi Jaakkola, and Amir Globerson. Learning efficiently with approximate inference via dual losses. ICML'10 | Lukas Muelleder |
| 23.06 | Rajhans Samdani and Dan Roth: Efficient Decomposed Learning for Structured Prediction" ICML'12 | Zoe Bylinovich |
| 30.06 | David Weiss and Ben Taskar: Structured Prediction Cascades AISTATS'10 | no session |
| 07.07 | Janardhan Rao Doppa, Alan Fern and Prasad Tadepalli: Output Space Search for Structured Prediction. ICML'12 | reading group (Julia Kreutzer) |
| 14.07 | Heng Yu, Liang Huang, Haitao Mi, and Kai Zhao: Max-Violation Perceptron and Forced Decoding for Scalable MT Training. EMNLP'13 Liang Huang, Suphan Fayong, and Yang Guo: Structured Perceptron with Inexact Search. NAACL'12 | Mayumi Ohta |
| 21.07 | Veselin Stoyanov, Alexander Ropson, and Jason Eisner: Empirical risk minimization of graphical model parameters given approximate inference, decoding, and model structure. AISTATS'11 Veselin Stoyanov and Jason Eisner: Minimum-risk training of approximate CRF-based NLP systems. NAACL'12 |
Literatur
Kevin Murphy "Machine Learning: A Probabilistic Perspective"
Daphne Koller and Nir Friedman "Probabilistic Graphical Models: Principles and Techniques"
Martin J. Wainwright and Michael I. Jordan "Graphical Models, Exponential Families, and Variational Inference"
