[mlpack] Neural Evolution Algorithms --- Week 6

[bang liu]

Dear all,

Last week, we finished the implementation of NEAT algorithm, which is the
key algorithm in our project. We also tested it by XOR testing case and it
passed. Besides, during the implementation of NEAT, we revised some basic
classes such as NeuronGene and Genome. The most important revision I think
is the introduce of "depth" for each neuron, which defines the priority, or
the sequence, that a neuron will be activated inside a genome. This concept
makes the calculation of neural network be quite clear and not ambiguous
even with looped or recurrent connection. Sequence of activating neurons
will be decided by sorting neurons with their depth in genome.

But this doesn't means that the NEAT algorithm is perfect now. Actually I
believe there are still some potential bugs existing and need to be
uncovered. Considering that the next algorithm, which is Hyper-NEAT, is
based on NEAT, and if we implemented a quite good NEAT algorithm, then
implementing Hyper-NEAT won't be too difficult. Thus, before starting
implementing Hyper-NEAT, this week I am planning to test NEAT using more
testing cases such as pole balancing problem and so on. More excited, I am
planning to try whether it works for playing games.

Best,
Bang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cc.gatech.edu/pipermail/mlpack/attachments/20160704/36340b48/attachment-0002.html>