Gneural Network - Tasks: task #14205, Implement NEAT for...
You are not allowed to post comments on this tracker with your current authentication level.
task #14205: Implement NEAT for genetic-algorithm training
Submitter: | Ray Dillinger <rayd> | ||
Submitted: | Sun 30 Oct 2016 05:48:21 PM UTC | ||
Should Start On: | Sun 30 Oct 2016 07:00:00 AM UTC | Should be Finished on: | Thu 02 Mar 2017 08:00:00 AM UTC |
Category: | None | Priority: | 5 - Normal |
Status: | None | Privacy: | Public |
Assigned to: | None | Percent Complete: | 0% |
Open/Closed: | Open | Effort: | 0.00 |
No files currently attached
Depends on the following items: None found
Items that depend on this one: None found
Carbon-Copy List
There are 0 votes so far. Votes easily highlight which items people would like to see resolved in priority, independently of the priority of the item set by tracker managers.
No changes have been made to this item
Powered by Savane 3.13-3230.
Corresponding source code
Neuro Evolution of Augmenting Topologies is a fairly simple, robust method of doing genetic algorithm training of neural networks, relying on a sequential record of when mutations to infer how to match up genomes of different lengths or topologies. It is certainly the 'canonical' normal algorithm for genetic training of neural networks. It kills simple problems dead.
That said the complexity of the solutions it can find is sharply limited by a uniform frequency of mutation, and it could be improved considerably with use of 'meta-genetics' that control mutation rates in different parts of the genome.
The problem is that uniform mutation rates indiscriminately destroy information as rapidly as it is created, which does not allow complex 'subassemblies' to ever stabilize under a mutation rate that allows the evolution of other complex 'subassemblies' - if the mutation rate is high enough to allow evolution of further complexity, it's high enough to destroy existing complexity.
Still, unmodified NEAT is good enough for simple 'reflex' type problems including lots of benchmarks such as steering, pole balancing, playing Mario, etc.