bigback Bayesian neural network simulator.
source code for bigback5.
information (not up to date).
I wrote bigback to develop and test the
ideas of my phd, for research purposes only. I used it to
make my winning entry in the prediction competition.
It can implement Bayesian methods for regression nets and
classifiers but is a research tool only.
Commercial users are asked to send a donation
to support my research
if they use this software.
Anyone interested in non-linear data modelling with
neural networks is encouraged to look into Gaussian processes.
Bigback has been used to create several Materials Modelling Algorithms.
See Materials Algorithms Project website
then go to Neural networks section
- macopt: A conjugate gradient optimizer that
I wrote, which only makes use of gradient information
and doesn't need a routine that returns the function value.
More information about
- Free Energy Code
tar. (For decoding and cryptanalysis. Please contact
me before taking this code.)
- Software for simulation of MN codes and GL codes (almost the
best error correcting codes in the world!)
tar. (c) David MacKay and Radford Neal.
see also Radford Neal's software
- A couple of codes for use with the above MN/GL software
can be found here:
tar. Or you can get individual files here
directory. There are three files for each code: A, Cn and G.
A is the parity check matrix, and G is the generator matrix.
Encyclopedia of sparse graph codes.
I have written tons of perl programs, which you can find
here. All supplied without warranty.
(Get perl here)
The Inference Group is supported by the Gatsby Foundation
and by a partnership award from IBM Zurich Research Laboratory
David J.C. MacKay
Site last modified Wed Nov 16 18:07:29 GMT 2005