Weekly summary 2016-10-15

Threads

• PMI - get some results!
• SoS - lectures 2 and 3
• DL: do experiments suggested in DL generalization
  • (*) NN learns DL. (Mon, Tue) - Partially written up. Running into trouble with some inequalities.
  • DL generalizations.
• Stability of SGD
  • @Elad: do local minima generalize? question statement (Tue)
• Graphical models reading notes (Thu)
• Alexa references
• Learn BCO (Tue, Wed)
• Papers (finish reading, summarize) - see BCO notes above and HMM notes (Tue, Wed)
  • LCLS10 A Contextual-Bandit Approach to Personalized News Article Recommendation
  • DHKK Efficient Optimal Learning for Contextual Bandits
  • SCH Unsupervised Part-Of-Speech Tagging with Anchor Hidden Markov Models
  • A spectral algorithm for learning Hidden Markov Models

Analyze Arora and Ge’s NMF algorithm in the presence of noise. Exactly how much noise can it tolerate?

• TODO Read this paper: [CFP16] Assessing significance in a Markov chain without mixing

Talk with Arora 10/13 (Thu)

• DL experiments
  • Convergence by initializing with samples?
  • Convergence with backprop?
  • • @Tengyu
• NNDL
• PMI - fix training problem, @Yingyu for advice on code
• Explaining backprop, the Network chin rule \(\pd fv = \sum_i \pd f{y_i} \pd{y_i}v\).
• Come up with a class of MDPs on exponential space that is interesting and tractable. Thoughts

Other

• Finite model theory (Thu)

[HMR16]

“correct initial state that generated \(y_t\)” Evolution is deterministic!