==== Presentation Slides ====
  * {{project:muri-13-tao.pdf|slides for MURI meeting, Nov. 22}}


==== Problem Formulation and Learning Methods ====
  * [[project:tensor_dp:draft_dec|Online Learning Method]]
  * [[project:tensor_dp:draft|Online Learning Method]]
  * [[project:tensor_dp:old_draft|Batch Learning Method]]


==== Properties of Matrix Norms  ====

  * [[project:tensor_dp:tracenorm_proof|Trace norm]]
  * [[project:tensor_dp:specturalnorm_proof|Spectral norm]]
  * [[project:tensor_dp:frobeniusnorm_proof|Frobenius norm]]





==== To Do ====
  * forgot to add core unigram features!
  * use integer feature code instead of string!
  * <del>Try 1/sqrt(k) learning rate</del>
  * <del>Try disabling lexical features</del>
  * <del>Evaluate based on seen/unseen words/arcs</del>
  * [[http://static.googleusercontent.com/media/research.google.com/en/us/pubs/archive/36898.pdf|Online Learning in the Manifold of Low-Rank Matrices]]

==== References ====

  * Martin Jaggi and Marek Sulovsky, **A Simple Algorithm for Nuclear Norm Regularized Problems**. ICML 2010 [[http://www.m8j.net/data/List/Files-149/fastRegNuclearNormOptimization.pdf|PDF]]
  * Elad Hazan, **Sparse Approximate Solutions to Semidefinite Programs**. LATIN'08 [[http://ie.technion.ac.il/~ehazan/papers/SparseSDP.pdf|PDF]]
  * [[wp>Matrix norm]]
  * Power method [[wp>Power method|Wiki]] [[http://ergodic.ugr.es/cphys/LECCIONES/FORTRAN/power_method.pdf|PDF]]