Weekly summary 2016-12-03

HMM/MDP project

• [ALA16] open problem, simple policy optimization - is it intractable? (Tue. started)
• [JKAL16] Contextual MDP paper (Tue. started)
• Go through policy gradient (Tue)
• Theory review
  • Ch. 3, 5 of FA
  • Ch. 6 Bertsekas
• Experiments
  • Program basic RL

Alexa/NLP experiments

• Get Yingyu’s word2vec, DL code up and running.
  • Try modifying with other modalities (sentiments, etc.)
• Lisa Lee’s thesis
• Parsing
  • Stanford parser
  • Try training NN
  • SyntaxNet
• Train word vectors with syntax trees.
• Read EMNLP papers.

Other experiments

• Neural net learns dictionaries: run experiments.
  • NNDL experiment.
  • Where is the cusp?

Learning

• EM algorithm
• Nonparametric/kernel methods (cf. BCO)
• Tensor decomposition and latent variable model
• MDP theory (see above)
• Summarize papers