Many neuroevolution methods evolve fixed-topology
Some methods evolve topologies in addition to
weights, but these usually have a bound on the complexity of
networks that can be evolved and begin evolution with random topologies.
This project is based on a
neuroevolution method called
NeuroEvolution of Augmenting Topologies (NEAT)
that can evolve networks of unbounded complexity from a minimal
The initial stage of research aims to
demonstrate that topology can be used to increase
the efficiency of search if it
minimizes the dimensionality of the weight space.
We performed several pole balancing experiments
that demonstrate that evolving topology using NEAT indeed
provides an advantage.
However, the research has a broader goal of showing
that evolving topologies is necessary to achieve 3
major goals of neuroevolution: (1) Continual coevolution: Successful competitive
coevolution can use the evolution of topologies to
continuously elaborate strategies. (2) Evolution of Adaptive Networks:
of topologies allows neuroevolution to evolve
adaptive networks with plastic synapses by designating
which connections should be adaptive and in what ways.
(3) Combining Expert Networks: Separate expert neural networks can be fused
through the evolution of connecting neurons between them.
Because we want to show that
growing structure is necessary to achieve these
goals, it is important that an efficient and principled method for
evolving topologies is available for experimentation.
NEAT provides just such an experimental platform.
NEAT is also an important contribution to
GAs because it shows how it is possible for evolution
to both optimize and complexify
making it possible to evolve
complex solutions over time, thereby strengthening the
- On the NEAT Method: Evolutionary Computation Journal Paper and
a Conference Paper (Best Paper Award Winner at GECCO-2002)
that goes into discussion of the way NEAT
A shorter second Conference Paper on improving NE efficiency with evolution
of topologies, emphasizing ablation studies.
- NEAT User Information:
NEAT Users Page includes a FAQ.
- On the benefits of Complexification: A Conference Paper on using the complexification of networks
to enhance the performance of competitive coevolution.
A more extensive and general
Journal Paper (new) about complexification.
The NEAT Demo Page
includes animated GIF movie clips of simulated robot
controllers coevolved using NEAT.
NEAT Software is available in C++, Java, and Matlab source code (see below)
- On Analyzing Results: Our papers on
complexification utilized a
method that we developed for monitoring progress in coevolution
called Dominance Tournament.
- Artificial Embryogeny: Journal
Paper on evolving structures that develop from a single
cell. A shorter symposium paper on combining complexification
with indirect encodings.
- Evolving Hebbian Networks: Conference Paper comparing the evolution of networks with dynamic and static synapses.