machine learning with discriminative methods lecture 18 – structured prediction cs 790-134 spring...

Download Machine Learning with Discriminative Methods Lecture 18 – Structured Prediction CS 790-134 Spring 2015 Alex Berg

If you can't read please download the document

Upload: lana-daughtery

Post on 15-Dec-2015

227 views

Category:

Documents


3 download

TRANSCRIPT

  • Slide 1

Machine Learning with Discriminative Methods Lecture 18 Structured Prediction CS 790-134 Spring 2015 Alex Berg Slide 2 Todays class Structured prediction Discuss next reading for deep learning Slide 3 Structure Prediction Some examples from Ben Taskar (UPenn and University of Washington) Slide 4 Handwriting Recognition brace Sequential structure xy Slide 5 Object Segmentation Spatial structure xy Slide 6 Natural Language Parsing The screen was a sea of red Recursive structure xy Slide 7 Bilingual Word Alignment What is the anticipated cost of collecting fees under the new proposal? En vertu des nouvelles propositions, quel est le cot prvu de perception des droits? xy What is the anticipated cost of collecting fees under the new proposal ? En vertu de les nouvelles propositions, quel est le cot prvu de perception de les droits ? Combinatorial structure Slide 8 Protein Structure and Disulfide Bridges Protein: 1IMT AVITGACERDLQCG KGTCCAVSLWIKSV RVCTPVGTSGEDCH PASHKIPFSGQRMH HTCPCAPNLACVQT SPKKFKCLSK Slide 9 Local Prediction Classify using local information Ignores correlations & constraints! breac Slide 10 Local Prediction building tree shrub ground Slide 11 Structured Prediction Use local information Exploit correlations breac Slide 12 Structured Prediction building tree shrub ground Slide 13 Formulating the problem There is a rich history, and we are skipping to a post-modern reductionist viewpoint (see first 5 sections of Nowozin reading) In particular we are avoiding a probabilistic formulation, but keep in mind that the following technique works for a probabilistic model, replacing armgax with maximum likelihood, and using sampling where appropriate Slide 14 Discriminative modeling for structured prediction We have already seen one example Multiclass classification! #is == #unique ys May be too flexible with large space of outputs Slide 15 Discriminative modeling for structured prediction Can we simplify f ? n