Last updated 4/21/2004
Development Status
Screenshot
Release Notes
Download
What is Backprop?
Backprop is a standalone multi-layer neural network simulator that is
based
upon the popular backpropagation learning algorithm. The goal of this
simulator
is to provide users with a friendly and easy to use environment for
experimenting
with backpropagation networks. To achieve this, I put a lot of effort
into
making the user interface give as much visual feedback as possible,
especially
during network training, as well as giving the user easy to use
interfaces
for changing the attributes of the network, such as learning rates,
momentum, and so forth. You can zoom in on the network graphically to
see weight values
in more detail, or zoom out in order to make visible larger, more
complicated
network architectures. You can speed up, or slow down, the rate at
which
error graphics and network state are updated during training.
Backprop displays activation and weights during training as they
change,
and allows the user to enable/disable/configure the use of momentum and
learning rate during training. You can also enable or disable a bias
term
to see what effect it has on convergence during training. Finally,
backprop
allows you to specify the use of sigmoid or htan activation functions.
Please refer to the release notes for more information and
instructions.
Version 0.9.5 was made available 12/27/2003. Please send your bug
reports and any feature requests to slogan621@gmail.com.
Version 0.9.5 for MacOS (New!!!) made available on 4/21/2004.
Features
introduced in 0.9.5 include:
- full XML support: network architecture, training exemplars, and
runtime data are all specified in XML
- ability to save modified network training settings
- ability to easily create and modify networks graphically (and
save them as XML)
Click
here.
Windows 98/NT/2000/XP Click
here for Backprop 0.9.5 (SIMTEL)
MacOS X 10.2 or greater Click here for Backprop
0.9.5
