Seth Holtquist

Seth Holtquist

Montgomery, Illinois

More

Haven't created any projects yet! 

Haven't backed any projects yet! 

No lab notes posted yet!

Hi Luis, thank you for your support! As it happens, the NEAT algorithm was one of the first avenues I explored in terms of neuroevolution. My approach is in several ways similar to or in part informed by NEAT. I am conducting several studies on various impacts of biological mimicry to inform my projects, but this current project focuses on the process of group evolution through a modified version of the Genetic Algorithm, in this case tested on the relatively well-understood image classification task. Specifically, the differences between NEAT and my approach are that the GA itself does not inform the weight matrices, but rather the specifics of the learning process in addition to architecture. NEAT and this project have a very similar end goal, but I go about it by selecting for architecture in terms of number of layers, neurons per layer, types of layers, etc as well as optimizer, error function, learning rate, and other learning hyperparameters in order to simulate the evolution of not just a fit individual, but a fit individual that learns easily and efficiently for a given problem.
Sep 13, 2018
Can neural networks increase in accuracy and efficiency by modeling evolution?
View comment
Thank you, Kieran! At the moment, I use a modified version of the genetic algorithm. I must confess I haven't heard of ant colony algorithms before; I'll have to look into those.
Sep 07, 2018
Can neural networks increase in accuracy and efficiency by modeling evolution?
View comment