Skip to content →

Trying to find time to develop bil, neural nets

I’ve two outstanding projects to take care of that have been on hold for awhile due to lack of time. The only thing that’s changed is that they can’t be on hold much longer.

bil was started in November 2008, and must be finished by August 7, 2009. I have yet to implement typing, functions, and arrays, and those three things are going to be a pain in the ass, especially given how little time I’ve got.

I’m also trying to complete a project determining the most efficient partially connected neural network to learn the next consecutive output of the sine function, given three consecutive, immediately prior points. To do this, I must evaluate all graphs between 3-1 (3 inputs, 1 output) and 3-3-1 (3 inputs, 3 hidden, 1 output, fully connected.) It’s probably that either 3-1 or a fully connected 3-3-1 will be best, but if there is a partially connected best, it will probably be publish worthy. The issue with this is finding a library that can implement designed partially connected graphs, of which in my experience, there are none. My Master’s dissertation will hopefully change this in that I plan to extend fann to support this prior to exploring grid computing of distributed neural nets. But for now, I have to implement this manually, which again requires time I don’t have.

After these, in the near future, I have to complete a project for visualizing data provided by sensors on a sensor network, and also formally submit my pre-proposal for my Master’s disseration.

Published in Uncategorized