|COOKIES: By using this website you agree that we can place Google Analytics Cookies on your device for performance monitoring.|
Structured Prediction using Linear Programming Relaxations
If you have a question about this talk, please contact Dr Daniel Roy.
Predicting structured objects such as parse trees or protein folds is often formulated as combinatorial optimization, where the goal is to find the most likely structure given the available evidence. Linear programming relaxations are a powerful tool for solving these optimization problems. Learning for structured prediction corresponds to inverse combinatorial optimization, finding parameters for the model such that for each of the data points, the optimal solution is the desired structure. This talk will survey algorithms and theory relating to learning for structured prediction using linear programming relaxations, as applied to dependency parsing in natural language processing, multi-label prediction, and protein side-chain placement.
Based on joint work with Michael Collins, Amir Globerson, Tommi Jaakkola, Terry Koo, Ofer Meshi, and Sasha Rush.
This talk is part of the Machine Learning @ CUED series.
This talk is included in these lists:
Note that ex-directory lists are not shown.
Other listsCreative Research at Museum of Archaeology & Anthropology Measuring National Well-Being – what matters to you? Simple Ideas that Change the World
Other talksRaymond and Beverley Sackler Distinguished Lecture Humans + Automation: Greater than the sum of the parts Hydrodynamic description of thin nematic films North Atlantic Oscillation: teleconnections, mechanisms and long range predictability Entanglement and Quantum Gate Processes in the One-Dimensional Quantum Harmonic Oscillator TBC