Genetic Programming Theory and Practice XVIII (Genetic and Evolutionary Computation)

Foreword
Preface
Contents
Contributors
1 Finding Simple Solutions to Multi-Task Visual Reinforcement Learning Problems with Tangled Program Graphs
	1.1 Introduction
	1.2 Tangled Program Graphs
		1.2.1 Learners
		1.2.2 Teams
		1.2.3 Graphs
		1.2.4 Memory
	1.3 Mechanisms for Accelerating TPG Evolution
		1.3.1 Rampant Mutation
		1.3.2 Multi-actions
	1.4 ViZDoom Subtask Selection and Performance Evaluation
	1.5 Empirical Methodology
		1.5.1 Task Domains
		1.5.2 Parameters
	1.6 Results
		1.6.1 Fitness
		1.6.2 Generalization
		1.6.3 Complexity
		1.6.4 Details of a RAPS Solution
	1.7 Conclusions
	References
2 Grammar-Based Vectorial Genetic Programming for Symbolic Regression
	2.1 Introduction
	2.2 State of the Art
		2.2.1 Vectorial Genetic Programming
		2.2.2 Grammar-Based Genetic Programming
		2.2.3 Feature Engineering and Feature Extraction
		2.2.4 Deep Learning
	2.3 Grammar-Based Vectorial Genetic Programming
		2.3.1 Vectorial Tree Interpretation
		2.3.2 Vectorial Symbolic Regression Grammar
	2.4 Experiment Setup
	2.5 Results
		2.5.1 Analysis Benchmarks Group A
		2.5.2 Analysis Benchmarks Group B
	2.6 Discussion and Next Steps
	References
3 Grammatical Evolution Mapping for Semantically-Constrained Genetic Programming
	3.1 Introduction
	3.2 Software Engineering Applications of Semantically–Constrained GP
		3.2.1 Automated Program Repair
		3.2.2 Automated Test Generation
		3.2.3 Program Synthesis
	3.3 Semantic Constraints in GP
		3.3.1 Strongly-Typed GP (STGP)
		3.3.2 Grammar-Guided GP (GGGP)
		3.3.3 Refined-Typed GP (RTGP)
	3.4 Correct-by-Construction Versus Generate-and-Validate
	3.5 Direct Versus Indirect Representations
	3.6 A Dynamic Grammar-Guided Mapping
		3.6.1 GE Mapping
		3.6.2 Semantic Filter of Valid Productions
		3.6.3 Dynamic and Depth-Aware Dynamic Approaches
	3.7 Evaluation
	3.8 Conclusions
	References
4 What Can Phylogenetic Metrics Tell us About Useful Diversity in Evolutionary Algorithms?
	4.1 Introduction
	4.2 Methods
		4.2.1 Selection Methods
		4.2.2 Problems
		4.2.3 Computational Substrates
		4.2.4 Other Parameters
		4.2.5 Phylogenetic Diversity Metrics
		4.2.6 Analysis Techniques
		4.2.7 Code Availability
	4.3 Results and Discussion
		4.3.1 Do Phylogenetic Metrics Provide Novel Information?
		4.3.2 Do Phylogenetic Metrics Predict Problem-Solving Success?
	4.4 Conclusion
	4.5 Author Contributions
	References
5 An Exploration of Exploration: Measuring the Ability of Lexicase Selection to Find Obscure Pathways to Optimality
	5.1 Introduction
	5.2 Exploration Diagnostic
	5.3 Lexicase Selection
		5.3.1 Epsilon Lexicase Selection
		5.3.2 Down-Sampled Lexicase Selection
		5.3.3 Cohort Lexicase Selection
		5.3.4 Novelty-Lexicase Selection
	5.4 Diagnosing the Exploratory Capacity of Lexicase Selection and Its Variants
		5.4.1 Lexicase Selection Out-Explores Tournament Selection
		5.4.2 The Exploratory Capacity of Lexicase Selection Degrades as We Increase Diagnostic Cardinality
		5.4.3 Increasing Population Size Can Improve Lexicase Selection\'s Exploratory Capacity
		5.4.4 Relaxing Lexicase Selection\'s Elitism Can Improve Exploration
		5.4.5 Down-Sampling Degrades Lexicase Selection\'s Exploratory Capacity
		5.4.6 Cohort Partitioning Degrades Lexicase Selection\'s Exploratory Capacity
		5.4.7 Cohort Lexicase Out-Explores Down-Sampled Lexicase
		5.4.8 Novelty Test Cases Degrade Lexicase Selection\'s Exploratory Capacity
	5.5 Conclusion
	5.6 Data and Software Availability
	References
6 Feature Discovery with Deep Learning Algebra Networks
	6.1 Introduction
	6.2 ARC Background
	6.3 Regression in Brief
	6.4 Classification in Brief
	6.5 Industrial Regression Classification
	6.6 Theoretical Test Problems—Classification
	6.7 Base Performance on the Theoretical Classification Problems
	6.8 Thin 2-Layer ARC Performance on the Theoretical Classification Problems
	6.9 Ultrathin 8-Layer ARC Performance on the Theoretical Classification Problems
	6.10 Wide 2-Layer ARC Performance on the Theoretical Classification Problems
	6.11 Wide 8-Layer ARC Performance on the Theoretical Classification Problems
	6.12 Conclusion
	References
7 Back to the Future—Revisiting OrdinalGP and Trustable Models After a Decade
	7.1 Introduction
	7.2 In the Beginning
		7.2.1 Model Complexity—Getting What You Measure
		7.2.2 ParetoGP—Simplicity and Accuracy
		7.2.3 Secondary and Alternating Objectives
		7.2.4 OrdinalGP—Failing Forward
		7.2.5 Ensembles—Trustable Models and Active Design-of-Experiments
		7.2.6 Data Balancing
	7.3 BalancedGP
		7.3.1 DataSubsetSize
		7.3.2 BalancedSample
		7.3.3 BalancedGP
	7.4 Ensembles
		7.4.1 Introduction to Ensembles
		7.4.2 Ensembles of the Future
	7.5 Conclusions
	References
8 Fitness First
	8.1 Introduction
	8.2 Faster Genetic Programming via Parallel Hardware
		8.2.1 Multiple CPU Cores
		8.2.2 Multiple Fitness Cases Simultaneously
		8.2.3 Fitness First
	8.3 Avoiding Effort Wasted on Poor Fitness Individuals
	8.4 Asymmetry of GP Subtree Crossover
		8.4.1 Last Child Inplace Dad-Less Crossover
	8.5 Efficiency of Memmove V. Memcpy
	8.6 Speed of Fitness First and Incremental Fitness
	8.7 Mathematical Model of Number of Parents
		8.7.1 Number of Parents Initially and in Diverse Populations
	8.8 Multi-threading Implementation Issues
		8.8.1 Idle Threads
		8.8.2 Future Work: Predicting Thread Execution Time
	8.9 Conclusions
	References
9 Designing Multiple ANNs with Evolutionary Development: Activity Dependence
	9.1 Introduction
	9.2 Multiple Problem Solving ANNs
	9.3 The Neuron Model
		9.3.1 Soma Program Inputs and Outputs
		9.3.2 Dendrite Program Inputs and Outputs
		9.3.3 Developing the Brain and Evaluating the Fitness
		9.3.4 Extracting Conventional ANNs from the Brain
		9.3.5 Activity Dependence
		9.3.6 Model Parameters
	9.4 Experiments
	9.5 Discussion and Further Work
	References
10 Evolving and Analyzing Modularity with GLEAM (Genetic Learning by Extraction and Absorption of Modules)
	10.1 Introduction
	10.2 Evolving Modules in Genetic Programming
	10.3 GLEAM
		10.3.1 Initializing the Library
		10.3.2 Referencing the Modules
		10.3.3 Updating the Library
	10.4 GLEAM as a Platform for Testing
	10.5 Experiments and Analysis
		10.5.1 Experimental Set-Up
		10.5.2 Using GLEAM to Evolve Modular Programs
		10.5.3 Using GLEAM as a Testing Platform
		10.5.4 Modular Usage in GLEAM
	10.6 Conclusions
	References
11 Evolution of the Semiconductor Industry, and the Start of X Law
	11.1 Introduction
	11.2 Human Knowledge Constraint
	11.3 Evolutionary Concepts
		11.3.1 What Evolutionary Components Can Be Applied to the Semiconductor Industry?
		11.3.2 What Else Does Evolution, and Economic Models Tell Us?
		11.3.3 How Can Ascension Occur?
		11.3.4 What About the Human Element?
	11.4 Final Discussion and Thoughts
		11.4.1 What are the Mechanisms for Continued Exponential Growth?
	11.5 Conclusion
	References
Index