Weekly summary 5-14-16

Machine learning

• Presentation on Language Games
  • TODO: Find some theoretical problems.
• NMF
  • Why can we use the regular pseudoinverse? (CONTINUE)
  • TODO: Search literature, talk to Andrej, Tengyu for ideas
  • TODO: Look at Freddy’s code and do some experiments.
• NLP learning
  • Read Mike Collins’s notes
  • Skimmed [CC14] on NMF.
  • TODO: Read book on NLP.
• Streaming PCA, Yuanzhi Li.
• TODO: Finish reviewing rest of BV
• TODO: Read lots, start keeping track of problems and techniques.

Coding

• Understood SoS relaxation approach.
• Started reading CSP refutation paper.
  • TODO: Read more of it, in particular, understand how to convert it to SoS language.
• Tried “covering with convex sets” but seems difficult.
• TODO: try approach with modulus of convexity.
• TODO: understand SDP extension approach.
• TODO: other approaches?

TCS

• Log-sobolev inequalities

TODO

• Review statistical physics.
• Review metric embeddings/geometric inequalities (MAT529) (@Kiran).
• Things left to summarize:
  • Convex geometry
  • Optimization (BV)
  • Optimization on manifolds (AMS)
• Make a summer reading list. (@Arora)
  • General TCS (randomized algorithms, approximation algorithms)
  • Probability (finish)
  • Advanced probability
    • ? Martingales
    • Concentration inequalities
    • Stochastic processes
    • Random matrices
  • Amit Singer’s course
  • Optimization (@Hazan for recommendations.)
  • ? ATaAP
  • Topics
    • Variational inference and MCMC
      • Volume estimation??
    • Graphical models

Other

• Olivier’s generals. TODO: Ask about CertiCoq.