MGL is a machine learning library focusing on Boltzmann Machines (BM), including Deep and Restricted Boltzmann Machines (DBM, RBM) and stacks of RBMs called Deep Belief Networks (DBN). It can turn a DBN or a DBM into a backprop network for fine-tuning. DBNs are state-of-the-art as of 2007. Backprop with gradient descent and conjugate gradient optimization is also available without ever having to touch DBNs.

For background on Boltzmann Machines see publications of Geoffrey Hinton.

MGL was written by Gabor Melis, with Ravenpack.

It can be found at;a=summary.

ASDF-install package (obsolete)