Weekly summary 2016-11-12

Summary

• Neural net learns dictionaries: run experiments. I found nothing unexpected: closeby initialization does converge. The farther away, the farther away the final \(A\) is. (But I think it should be a _/ shape—where is the cusp?)
  • Are the different dictionaries related by a linear transformation? It seems not.
• Simons meeting—spanning trees, bandit convex optimization.
  • Can we use kernel methods for MDP’s?
• Meeting with Arora
  • Learn about continuous HMM models.
  • Improve NLP.