Hello

About me

• Multilayer deterministic feedforward neural networks with unusual ingredients (see thesis).
• Multilayer neural networks, using Restricted Boltzmann Machines.
• Using artificial intelligence as a tool (as opposed to purpose) in programs - that's interesting because it's a useful tool, and sometimes it can do some of the boring work for me.

• The theory and practice of programming (e.g. compilers and interpreters, algorithm design, programming languagues, meta programming, and programming tricks) - that's interesting because I program a lot. See also my programming page.
• The study of logic and the principles of mathematics. I enjoy studying this, and I've taken some excellent courses in this field.
• Probability theory - especially its paradoxes (like the paradox of the two envelopes).

Publications

• Optimizing neural networks that generate images
  PhD thesis, 2014
  [pdf]
  
  
• Gnumpy: an easy way to use GPU boards in Python
  Tijmen Tieleman
  Technical report UTML-TR-2010-002
  [ps] [pdf] [bibtex] [software]
  
  
• On the convergence properties of Contrastive Divergence
  Ilya Sutskever and Tijmen Tieleman
  In: AI STATS 2010
  [pdf] [bibtex]
  
  
• Using Fast Weights to Improve Persistent Contrastive Divergence
  Tijmen Tieleman and Geoff Hinton
  In: International Conference on Machine Learning (ICML) 2009
  [ps.gz] [pdf] [bibtex] [recorded video of the talk] [ppt without video files] [ppt with video files, 12 MB] [MNORB dataset in .tgz format]
  
  
• Training Restricted Boltzmann Machines using Approximations to the Likelihood Gradient
  Tijmen Tieleman
  In: International Conference on Machine Learning (ICML) 2008
  [ps.gz] [pdf] [bibtex] [ppt] [recorded video of the talk]
  
  
• Some investigations into energy-based models
  Tijmen Tieleman
  Master's thesis
  [ps.gz] [pdf] [bibtex]

Software & tools