I thought I would make available some presentation slides I have created recently. I will be giving a presentation in my course for artificial neural networks tomorrow on cascade correlation.Download Slides
For those unacquainted with cascade correlation it is a dynamic neural network which as it is processing determines its own topology. Networks begin with only input and output layers, and as necessary hidden units are introduced to the network.
The process requires to back propagation of error signals and often results in very unique solutions to problems. Cascade correlation's learning process is many times quicker than traditional back propagation, but in smaller problems its accuracy suffers. However, due to its dynamical creation, it tends to scale to much larger problems more efficiently while maintaining its break neck speeds of classification.
It's something I would like to experiment with over the summer once things calm down a bit. There are a few unexplored areas of study involved which I think potential papers are lurking.
0 comments:
Post a Comment