Papers and Links 2013-06-21


  • Spectral Learning Algorithms for Natural Language Processing. Shay Cohen, Michael Collins, Dean Foster, Karl Stratos, Lyle Ungar [PDF] Spectral learning is an alternative method for learning model parameters, the advantage seems to be that it provides theoretical guarantees that EM does not.
  • Upcoming Interspeech paper about discriminatively training individual WFSTs that are part of integrated recognition cascade. One point I'm not totally clear about is the choice of features.
  • Graham Neubig has a great set of slides about Non-parametric Bayesian Statistics
  • Massive Languge model built with KenLM on the ClueWeb data.
  • A couple of upcoming Microsoft papers on Convulution DNNs and Segmental DNNs for speech recognition. Interesting, but seem to be lower than than Alex Graves' results on Timit.
  • Nice paper comparing OpenCL and CUDA.
  • ICML2013 by Vincent Vanhoucke about recent advances in Acoustic Modeling and Deep Learning


  • PyBindGen Alternate tool for wrapping C/C++ for interoping with Python. (Thanks Joe)
  • Jekyll Bootstrap looks like by far the easiest way to get a static site generator running and deployed on github.


  • Vim regex to convert current line to title casing :s/\<\(\w\)\(\w*\)\>/\u\1\L\2/g
  • Just found that is0646 added alternative tokens for C/C++ logical and bitwise operations. Looks a lot like python
  • MITLM defatuls to trigram for all operations including evaluation and interpolation, remember to manually set the -o flag.
  • If cut and paste stops working between Windows remote desktop and the OSX client. Kill the rdclip.exe and restart it on the server.