Transcript Evolutionary Computation for Creativity and

Evolutionary Computation for
Creativity and Intelligence
{
By Lauren Gillespie, Gabby Gonzalez, and Alex Rollins
Neuroevolution: an overview

Neural Networks - Brains
Nodes – Neurons
 Links – Synapses

Genotype - Phenotype
 Evolutionary Algorithm - Abstraction of evolution

Asexual reproduction (mutation)
 Sexual reproduction (crossover)
 Survival of the fittest (selection)

Neural Networks + Evolution = Neuroevolution
 Neuroevolution of Augmenting Topologies (NEAT)

Compositional Pattern Producing
Networks (CPPN)
(R, G, B)
(H, S, B)
H
X
S
Y
B
D
Actual Network
Bias
(X, Y, D, Bias)
Picbreeder Demo
{
Evolved Behavior
Three blue predators
Two green prey
World is torus shaped
(wraps around)
Predators evolved to
catch the prey
Average Number of Prey Caught
Fitness Score
Generation
Fitness Functions
Prey
Predator
Minimize
Distance
Minimize
Prey Survival
Time
Maximize Number
of Prey Caught
Coevolution
Population
Evolved
Predator Vs
Static Prey
Grid World
Cooperative
Coevolution of
Predators
Static
Controller
Prey Agent
Network
Predator Agent
Competitive
Coevolution
(Homogenous Teams)
Competitive and
Cooperative
Coevolution
Checks each possible
move it can make.
{
Uses neural network to
rate each move
Picks the best action
(maximum utility)
Evolved Tetris Player
Tetris Features
Based on features from Bertsekas, D. P. and Tsitsiklis, J. N. (1996). Neuro-Dynamic Programming. Athena Scientific
Fitness Score
Tetris movie here too
Game Score

Evolved Tetris Player
Generation
Hybercube-based NEAT (HyperNEAT)
Stanley, Kenneth O.; B. D'Ambrosio, David; Gauci, Jason (2009). "A Hypercube-Based Encoding for Evolving LargeScale Neural Networks". Artificial Life 15: 185–212.
Tetris with Raw Feature Inputs (HyperNEAT)
Utility
Classic Game “Doom” (VizDoom)
- Use raw screen pixels to make decisions
- Try using single row in initial experiments
Using Full Screen (HyperNEAT)

Dr. Jacob Schrum

Southwestern University

HHMI-Inquiry Initiative
Howard Hughes
Medical Institute
Questions?
Auxiliary slides
How Neuroevolution works
Neuroevolution
1.
2.
3.
4.
Different networks
encode different
phenotypes
Phenotypes compete in
task
Networks evaluated on
phenotype score
Mutation and crossover
modify best networks
Biological evolution
1.
2.
3.
4.
A population of creatures
has slightly different traits
based on DNA differences
Environment exerts
pressure on population
Natural selection occurs,
fittest members survive
Survivors reproduce both
sexually and asexually
Neuro-Evolution of Augmenting
Topologies (NEAT)




Evolutionary Algorithm
Complex agents evolved
from simple networks
Complexity built up via
mutation and mating
Add a little more detail in
case of questions
The Network and the Sensors
Inputs and Outputs for
a Single Agent
Network for a Single Agent
Do Nothing
Multi-Objective Optimization

Imagine game with
two objectives:


Minimize Distance
Maximize Number of
Prey Caught
A dominates B if and
only if A is strictly
better in one objective
and at least as good
in others
 Population of points
not dominated are
best: Pareto Front
