Preface
Acknowledgments
Contents
Contributors
1 TPOT2: A New Graph-Based Implementation of the Tree-Based Pipeline Optimization Tool for Automated Machine Learning
1.1 Introduction
1.2 Related Work
1.3 Evolutionary Algorithm
1.4 GraphPipelineIndividual Representation
1.4.1 Mutation
1.4.2 Crossover
1.5 TPOT2 API
1.5.1 TPOTEstimator
1.5.2 Ensembling
1.6 Experiment Set-Up
1.6.1 TPOT1 Versus TPOT2
1.7 Results and Discussion
1.8 Conclusions
1.8.1 Future Work
References
2 Analysis of a Pairwise Dominance Coevolutionary Algorithm with Spatial Topology
2.1 Introduction
2.2 Preliminaries
2.2.1 Coevolutionary Algorithms
2.2.2 Spatial Topologies for PDCoEA
2.2.3 Error Thresholds
2.2.4 Problems
2.2.5 MaximinHillβA Problem for Error Thresholds
2.3 Experimental Methodology
2.4 Experiments
2.4.1 Setup
2.4.2 Spatial Topology PDCoEA
2.4.3 Payoff and Genotypic Diversity
2.4.4 Error Threshold in STPDCoEA
2.5 Related Work
2.6 Conclusion
References
3 Accelerating Image Analysis Research with Active Learning Techniques in Genetic Programming
3.1 Introduction
3.2 Data Sets
3.2.1 KOMATSUNA
3.2.2 Cell Classification
3.3 Active Learning
3.4 AL-GP Applied to Decision Tree GP
3.4.1 Decision Tree GP (DT-GP)
3.4.2 Active Learning Implementation
3.4.3 KOMATSUNA Multi-image Results
3.4.4 KOMATSUNA Single-Image Results
3.4.5 Cell Classification
3.5 AL-GP Applied to SEE-Segment
3.5.1 SEE-Segment
3.5.2 AL Implementation for SEE-Segment
3.5.3 KOMATSUNA Results
3.6 Conclusions
References
4 How the Combinatorics of Neutral Spaces Leads Genetic Programming to Discover Simple Solutions
4.1 Introduction
4.2 Related Work
4.2.1 I/O Systems
4.2.2 RNA Studies
4.2.3 GP on Boolean Functions
4.2.4 Neutral Networks
4.2.5 Our Earlier Work
4.3 Genotypes, Phenotypes, Behavior, Fitness
4.3.1 Discrimination of Genotypes and Phenotypes
4.3.2 The Difference of Structural and Semantic Neutrality
4.4 Methods
4.4.1 Linear Genetic Programming
4.4.2 Boolean Function Programs/Circuits
4.4.3 Visualization Method
4.5 The Role of Neutrality
4.5.1 Longer Programs
4.5.2 A New Fitness Function
4.6 Results
4.6.1 A Comparison of Success Rates
4.6.2 Comparison of Search Trajectory Networks for Three Targets
4.6.3 Simpler Solutions
4.7 Discussion and Future Work
References
5 The Impact of Step Limits on Generalization and Stability in Software Synthesis
5.1 Introduction
5.2 Background
5.2.1 The Push Language and Interpreter
5.2.2 Step Limits and Infinite Loops
5.2.3 Success, Generalization, and Stability
5.3 Methodology and Experimental Design
5.4 Results
5.4.1 Last Index of Zero
5.4.2 Fuel Cost
5.4.3 Middle Character
5.4.4 GCD
5.5 Discussion
5.5.1 Stability of Evolved Programs
5.5.2 Stability and (mis)match with Instruction Set
5.5.3 Finding Additional Generalizing Solutions
5.5.4 Saving Computational Effort
5.6 Future Work
5.7 Conclusions
References
6 Genetic Programming Techniques for Glucose Prediction in People with Diabetes
6.1 Introduction
6.2 The Problem of Glucose Management
6.3 Background
6.3.1 Grammatical Evolution for Glucose Prediction
6.3.2 Recent Techniques for Glucose Prediction Based on Grammatical Evolution
6.4 Proposed Framework for Glucose Control
6.4.1 Framework Description
6.4.2 Experimental Results
6.5 Conclusions
References
7 Methods for Rich Phylogenetic Inference Over Distributed Sexual Populations
7.1 Introduction
7.2 Methods
7.2.1 Genome Instrumentation
7.2.2 Genealogical Inference
7.2.3 Population Size Inference
7.2.4 Positive Selection Inference
7.2.5 Software and Data
7.3 Results and Discussion
7.3.1 Genealogical Inference
7.3.2 Population Size Inference
7.3.3 Positive Selection Inference
7.4 Conclusion
References
8 A Melting Pot of Evolution and Learning
8.1 Introduction
8.2 Machine Learning
8.2.1 Binary and Multinomial Classification Through Evolutionary Symbolic Regression ch8Sipper2022esr
8.2.2 Classy Ensemble: A Novel Ensemble Algorithm for Classification ch8sipper2022classy
8.2.3 EC-KitY: Evolutionary Computation Tool Kit in Python ch8eckity2023
8.3 Deep Learning
8.3.1 Evolution of Activation Functions for Deep Learning-Based Image Classification ch8Lapid2022
8.3.2 Adaptive Combination of a Genetic Algorithm and Novelty Search for Deep Neuroevolution ch8SegalS22
8.4 Adversarial Deep Learning
8.4.1 An Evolutionary, Gradient-Free, Query-Efficient, Black-Box Algorithm for Generating Adversarial Instances in Deep Networks ch8Lapid2022Query
8.4.2 Foiling Explanations in Deep Neural Networks ch8Vitrack2023
8.4.3 Patch of Invisibility: Naturalistic Black-Box Adversarial Attacks on Object Detectors ch8Lapid2023
8.5 Concluding Remark
References
9 Particularity
9.1 Overview
9.2 Lexicase
9.3 Variance
9.4 Epsilon
9.5 Batched
9.6 Downsampled
9.7 Informed
9.8 Weighted
9.9 Gradient
9.10 Plexicase
9.11 Hidden
9.12 Living
9.13 Honor
References
10 The OpenELM Library: Leveraging Progress in Language Models for Novel Evolutionary Algorithms
10.1 Introduction
10.2 Background: Evolution and LLMs
10.3 OpenELM Evolutionary Algorithms
10.4 Language Models as Evolutionary Operators
10.4.1 Diff Models
10.4.2 LMX: Language Model Crossover
10.5 Engineering Challenges
10.5.1 OpenELM Inference Optimizations
10.5.2 Execution of Generated Code
10.6 OpenELM Domains
10.6.1 Sodarace
10.6.2 Image Generation
10.6.3 Prompts
10.6.4 Programming Puzzles
10.7 Discussion
10.8 Conclusion
References
11 GP for Continuous Control: Teacher or Learner? The Case of Simulated Modular Soft Robots
11.1 Introduction
11.2 Related Works
11.3 Background: Simulated Voxel-Based Soft Robots
11.3.1 VSR Morphology
11.3.2 VSR Controller
11.4 Evolutionary Optimization of VSR Controllers
11.4.1 Multi-layer Perceptron Optimized with a Genetic Algorithm
11.4.2 Array of Regression Trees Optimized with GP
11.4.3 Regression Graphs Optimized with GraphEA
11.5 Experiments and Results
11.5.1 Direct Evolution of the Controller
11.5.2 Offline Imitation Learning
11.6 Discussion
11.7 Concluding Remarks
References
12 Shape-constrained Symbolic Regression: Real-World Applications in Magnetization, Extrusion and Data Validation
12.1 Introduction
12.2 Related Work
12.3 Shape-constrained Symbolic Regression
12.3.1 Interaction Transformation Evolutionary Algorithm
12.4 Shape Constraint Handling
12.4.1 Single-Objective Approach
12.4.2 Multi-objective Approach
12.4.3 Feasible-Infeasible Two-Population Approach
12.5 Constraint Evaluation
12.5.1 Optimistic Approach
12.5.2 Pessimistic Approach
12.6 Real World Problems
12.6.1 Twin-Screw Extruder Modeling
12.6.2 Data Validation for Industrial Friction Performance Measurements
12.6.3 Magnetization Curves
12.7 Conclusion
References
13 Phylogeny-Informed Fitness Estimation for Test-Based Parent Selection
13.1 Introduction
13.2 Phylogeny-Informed Fitness Estimation
13.2.1 Phylogeny Tracking
13.3 Methods
13.3.1 Lexicase Selection
13.3.2 Diagnostic Experiments
13.3.3 Genetic Programming Experiments
13.3.4 Statistical Analyses
13.3.5 Software and Data Availability
13.4 Results and Discussion
13.4.1 Phylogeny-Informed Estimation Reduces Diversity Loss Caused by Subsampling
13.4.2 Phylogeny-Informed Estimation Improves Poor Exploration Caused by Down-Sampling
13.4.3 Phylogeny-Informed Estimation Can Enable Extreme Subsampling for Some Genetic Programming Problems
13.5 Conclusion
References
14 Origami: (un)folding the Abstraction of Recursion Schemes for Program Synthesis
14.1 Introduction
14.2 Recursion Schemes
14.2.1 Fixed Point of a Linked List
14.2.2 Functor Algebra
14.2.3 Well-Known Recursion Schemes
14.3 Origami
14.3.1 How to Choose a Template
14.3.2 Jokers to the Right: Catamorphism
14.3.3 When You Started Off with Nothing: Anamorphism
14.3.4 Stuck in the Middle with You: Hylomorphism
14.3.5 Clowns to the Left of Me: Accumorphism
14.4 Preliminary Results
14.5 Discussion and Final Remarks
References
15 Reachability Analysis for Lexicase Selection via Community Assembly Graphs
15.1 Introduction
15.2 Approach
15.2.1 Community Assembly Graphs
15.2.2 Calculating Stability
15.2.3 Assumptions
15.2.4 Reachability Analysis
15.3 Background
15.3.1 Lexicase Selection
15.3.2 Community Assembly Graphs
15.4 Proof of Concept in NK Landscapes
15.4.1 Methods
15.4.2 Results
15.5 Proof of Concept in Genetic Programming
15.5.1 Methods
15.5.2 Results
15.6 Conclusion
References
16 Let's Evolve Intelligence, Not Solutions
16.1 Introduction
16.2 What Should We Strive For?
16.3 What Assumptions Are Limiting Us?
16.3.1 Posit#1: Impossible to Engineer Intelligence
16.3.2 Posit #2: No Occam's Razor for Intelligence
16.3.3 Posit #3: Intelligence Is Grounded
16.3.4 Posit #4: Intelligence Is Transferable
16.3.5 Posit #5: Intelligence Is Intrinsically Self-reinforcing
16.4 What Do We Need?
16.4.1 A Caveat: Intelligence == Process and/or Intelligence == Capabilities and/or Intelligence == Individual(s)
16.4.2 The World
16.4.3 The Drivers
16.4.4 Models of Understanding
16.4.5 Process of Intelligence Self-Reinforcement
16.5 How Should We Approach It?
16.5.1 Revisiting Reproducibility
16.5.2 Back to the Intelligence Function
16.5.3 Genetic Programming of Intelligence
16.6 Conclusions
References
Appendix Index
Index