Structure in the flow

This is a “weekly report” for the lab I study in, mostly intended for other lab members. See the first one for further explanations.

Readings

• I’ve read the part of the following paper that concerns their approach (not the experiments). I think I get the idea, but I need to think about what James proposed, ie. how this can fit with Weston et al.’s criteria.
  • Koray Kavukcuoglu, Marc’Aurelio Ranzato, Rob Fergus and Yann LeCun: Learning Invariant Features through Topographic Filter Maps, Proc. International Conference on Computer Vision and Pattern Recognition (CVPR’09), IEEE, 2009
• Razvan has sent me some Theano-based code he’s written to classify MNIST digits with a deep net of RBMs. I haven’t looked into the details (the code it depends on), but I understand pretty well what’s there.
• I spent the best part of monday reading about dynamic systems, mostly online resources suggested by Yoshua and some complements from Wikipedia.

General course/curriculum work

• For the NLP class assignment, we’ve done quite a bit of cleanup/debugging, and the results are now much better. Report is partly done, and we’ve now got to run the experiments cleanly, but that’s a pretty linear process since most tests have already been run in some form or another.
• In the ML class, there was the midterm exam, and the second homework is now available (concerns linear regression models).

Plan for next week (outside course work)

• I need to actually try running some non-toy Theano code, probably Razvan’s at first.
• I want to take a look at Guillaume’s LeDeepNet code, and hopefully come up with a project plan on how to glue things together for my ML class term project.
• Resume reading Learning deep architectures for AI (I tried to this week, but time was scarce).