Due Dates

Paper selection: Tuesday, March 20. Email cis620@seas.

Slides: Two days before your talk. Meet with instructors.

Presentation: In Class.

• Tuesday, March 27.
• Thursday, March 29.
• Tuesday, April 3.
• Thursday, April 5.
• Tuesday, April 10.
• Thursday, April 12.
• Tuesday, April 17.

Guidelines

Themes

• Approximate Inference: Presenting one of the many techniques not taught in class.
• Semisupervised and Unsupervised Learning: Learning with indirect feedback in the context of graphical model is challenging and an open field of research.
• Structured Prediction: Learning models for making multiple, related predictions at once. Representative examples include pixel-labeling we discussed in class or parsing a sentence, where the input is a sentence, and the output is a parse tree of a sentence.
• Application: The tools presented in class are used in many domains: natural language processing, speech analysis, image processing and others. Present a specific application with the approach used to solve the problem. Why did the methods work well?

Evaluation

• Content: Is a good range of information included (not too obvious nor too specialized)? Are the main points covered?
• Delivery: Is the presentation clear and concise? Well practiced?
• Organization: Is the talk well-structured? Easy to follow?

Some example papers:

• Approximate Inference:
  • Expectation Propagation for approximate Bayesian inference
  • Power EP
  • Approximate Inference Using Planar Graph Decomposition
  • Approximate Inference Using Conditional Entropy Decompositions
  • Markov Chain Monte Carlo (many papers)
  • Log-determinant relaxation for approximate inference in discrete Markov random fields
• Unsupervised and Semisupervised Learning:
  • Maximum Margin Semi-Supervised Learning for Structured Variables
  • Gene Expression Time Course Clustering with Countably Infinite Hidden Markov Models
  • Discriminative unsupervised learning of structured predictors
• Structured Prediction:
  • Discriminative Training Methods for Hidden Markov Models: Theory and Experiments with Perceptron Algorithms
  • Online Large-Margin Training of Dependency Parsers
  • Semi-Markov Conditional Random Fields for Information Extraction
  • Max Margin Markov Networks
  • Learning as Search Optimization: Approximate Large Margin Methods for Structured Prediction
• Applications:
  • Phoneme Alignment Based on Discriminative Learning
  • Graphical Models and Automatic Speech Recognition
  • Trigger-based Language Modeling using a Loss-sensitive Perceptron Algorithm
  • Learning to Map Sentences to Logical Form: Structured Classification with Probabilistic Categorial Grammars
  • Shallow Parsing with Conditional Random Fields
  • CC Prediction with Graphical Models
  • A Conditional Random Field for Discriminatively-trained Finite-state String Edit Distance