Upload
mohammad-shaker
View
636
Download
2
Embed Size (px)
Citation preview
Adaptive Games Content Generation
“Mario”
Mohammad ShakerDepartment of Artificial Intelligence
IT University of DamascusSeminar of Artificial Neural Networks
ZGTR
Outline
• Readings
• Motivation
• The proposed approach
• Experiments
• ANN Implementation
• Results
• Conclusion
Readings• Towards Automatic Personalized Content
Generation for Platform Games Noor Shaker, Georgios N. Yannakakis, Member, IEEE, and Julian Togelius,
Member, IEEE
• Feature Analysis for Modeling Game Content
Quality Noor Shaker, Georgios N. Yannakakis, Member, IEEE, and Julian Togelius,
Member, IEEE
Motivation
Motivation
Motivation
Motivation
Motivation
Motivation
Motivation
The Big Picture
The Big Picture
Game Player
The Big Picture
Game Player
The Big Picture
Game Player
Player ExperienceModel
The Big Picture
Game Player
Player ExperienceModel
Game Adaptation
The Big Picture
Game Player
Player ExperienceModel
Game Adaptation
The Big Picture
Game Player
Player ExperienceModel
Game Adaptation
The Big Picture
Game Player
Player ExperienceModel
Game Adaptation
The Game
The Game
Open Questions! Session period? (frequency of adaptation)
The most useful information about game content?
Game aspects with major affect on player
experience?
Open Questions! Session period? (frequency of adaptation)
The most useful information about game content?
Game aspects with major affect on player
experience?
Approach
Design
Approach
Design CollectData
Approach
Design CollectData
ModelPlayer’s Emotion
Data Collection 40 small levels
(one-third of usual size)
600 game pairs
Features Six controllable features
Players preferences of engagement
Data Collection 40 small levels
(one-third of usual size)
600 game pairs
Features Six controllable features
Players preferences of engagement
Data Collection - Controllable Features
number of gaps
average width of gaps
number of enemies
number of powerups
number of boxes
Enemies placement Around horizontal boxes
Around gaps
Random placement
Experiments
Experiment 1 How long the game session should be in order to be
able to extract useful information?
Experiment 1 How long the game session should be in order to be
able to extract useful information?
Segmentation
Segmentation
Content-Driven Preference Learning
• It’s the use of genetic algorithms to evolve the
weight of neural networks to learn preference data.
Content-Driven Preference Learning
Levels
• It’s the use of genetic algorithms to evolve the
weight of neural networks to learn preference data.
Content-Driven Preference Learning
Levels Segmentation
• It’s the use of genetic algorithms to evolve the
weight of neural networks to learn preference data.
Content-Driven Preference Learning
LevelsFeature
extractionSegmentation
• It’s the use of genetic algorithms to evolve the
weight of neural networks to learn preference data.
Content-Driven Preference Learning
NeuroEvolutionarypreference
learningLevels
Feature extraction
Segmentation
• It’s the use of genetic algorithms to evolve the
weight of neural networks to learn preference data.
Content-Driven Preference Learning
NeuroEvolutionarypreference
learningLevels
Player’sEngagement
Feature extraction
Segmentation
• It’s the use of genetic algorithms to evolve the
weight of neural networks to learn preference data.
NeuroEvolutionarypreference
learning
Feature extraction
Content-Driven Preference Learning
Feature extraction
Feature extraction
Feature extraction
NeuroEvolutionarypreference
learning
NeuroEvolutionarypreference
learning
NeuroEvolutionarypreference
learning
Experiment 2 How can we extract the most useful information
about game content?
Experiment 2 How can we extract the most useful information
about game content?
Game Content Representation
Statistical features
Sequences
Game Content Representation
Statistical features Six controllable features
Used for level generation
Sequences
Game Content Representation
Statistical features Six controllable features
Used for level generation
Sequences Numbers representing different types of game content
o Platform structure, S
o Enemies placement, Ep
o Enemies and items placement, D
Sequence Mining
Sequence Mining
Sequence Mining-SPADE
SPADE
Frequent
Subseq.
occurrences
40 levels seq.
Content-Driven Preference Learning
ANN-
NeuroEvolutionary
Preference
Learning
Statisticalfeatures
Player’s Engagement
ANN-
NeuroEvolutionary
Preference
Learning
Sequentialfeatures
Player’s Engagement
Experiment 3 What are the game aspects that have the major
affect on player experience?
Experiment 3 What are the game aspects that have the major
affect on player experience?
Content-Driven Preference Learning
ANN-
NeuroEvolutionary
Preference
Learning
Sequentialfeatures
Player’s Engagement
Statisticalfeatures
Featureselection
ANN Implementation
ANN Implementation• Multilayer perceptrons (MLPs)
o ANN inputs
• Controllable features
• Sequences as features
o ANN output
• Value of the engagement preference
ANN Training• Genetic algorithms (GAs)
o No prescribed target outputs
ANN Training• Genetic algorithms (GAs)
o No prescribed target outputs
• How it works?
ANN Training• Genetic algorithms (GAs)
o No prescribed target outputs
• How it works?
players’ reported
emotional preferences
magnitude of corresponding model (ANN)
output
ANN Training• Genetic algorithms (GAs)
o No prescribed target outputs
• How it works?
players’ reported
emotional preferences
magnitude of corresponding model (ANN)
output-
ANN Implementation
ANN Implementation
SF
CF
ANN Implementation
SF
CF
Optimizing Neural Networks Topologies
• 2 hidden layers (Max.)
Optimizing Neural Networks Topologies
• 2 hidden layers (Max.)
• Multiple experiments 1 hidden layer, Adding two neurons at each step
2 neurons - 8 neurons
Optimizing Neural Networks Topologies
• 2 hidden layers (Max.)
• Multiple experiments 1 hidden layer, Adding two neurons at each step
2 neurons - 8 neurons
2 hidden layers, Adding two neurons at each step
1st Hidden layer
2 neurons - 10 neurons
2nd Hidden layer
2 neurons - 8 neurons
Optimizing Neural Networks Topologies
• 2 hidden layers (Max.)
• Multiple experiments 1 hidden layer, Adding two neurons at each step
2 neurons - 8 neurons
2 hidden layers, Adding two neurons at each step
1st Hidden layer
2 neurons - 10 neurons
2nd Hidden layer
2 neurons - 8 neurons
ANN Adaptation
ANN Implementation
SF
CF
ANN Adaptation
SF
CF
Prediction ofplayer’s emotion
ANN Adaptation
SF
CF
Prediction ofplayer’s emotion
Gaps #: 4-10Gaps width: 10-30Gaps placement: 0-1Switch:0-1
ANN Adaptation
SF
CF
Prediction ofplayer’s emotion
Exhaustive search
Gaps #: 4-10Gaps width: 10-30Gaps placement: 0-1Switch:0-1
ANN Adaptation
SF
CF
Prediction ofplayer’s emotion
Exhaustive search
Gaps #: 4-10Gaps width: 10-30Gaps placement: 0-1Switch:0-1
ANN Adaptation
level1 level2
Adapt
level20
Adapt Adapt
level21 level50
Adapt
ANN-
NeuroEvolutionary
Preference
Learning
Sequentialfeatures
Player’s Engagement
Statisticalfeatures
Featureselection
Neural Networks Input Representation
Game Content Representation
Sequentialfeatures
Statisticalfeatures
Featureselection
Game Content Representation
Game Content Representation
The best-performing MLP models evaluated on occurrences
of frequent subsequences of length three extracted from the 40 levels
The topology and performance of the best MLP models evaluated on full and
partial information about game content. the MLP performance presented is the
average performance over 20 runs.
MLPs Performance on Full Information about Game Content
Results
The performance and topologies of MLP models evaluated on full and partial
information of game content using statistics from the game window and from two
and three segments to which the window has been divided. The performance
presented is the average over five runs.
Content-Driven Preference Learning
ANN-
NeuroEvolutionary
Preference
Learning
Sequentialfeatures
Player’s Engagement
Statisticalfeatures
Featureselection
Conclusion
Combining both sequential and statistical features
gives better results in predicting players' reported
emotional state.
Partitioning the level causes a significant decrease
(p < 0.05) in the accuracy of predicting player’s reported engagement. This suggests that there
might be information loss because of decomposing
the data and that this loss causes a performance
decrease.
Multiple perspectives can be done in reference to
this study which is already going on!
Thank you!