The vital functions of life are driven by proteins — long folded strings of amino acids. Millions of proteins exist in nature, all optimized for a certain set of operating conditions. One particularly important trait of a protein is thermostability, or the tendency to maintain form and function at high temperatures. Protein thermostability is important for both biological organisms and industrial enzymatic processes that utilize them. This work proposes a novel hybrid EC approach, employing the EAs Simulating Molecular Evolution (EASME) software, for evolving thermostable variants of proteins with minimal computational overhead and customized problem-specific fitness functions. A variation of NSGA-II with a novel recombination/mutation interplay was designed to tackle the specific challenges of working with proteins. The results of this work, which computationally improved the thermostability of a protein more effectively than a state of the art machine learning model from 2025, demonstrate that EC may have been historically undervalued in the realm of protein design. The pipeline designed for this work could be applied to numerous other proteins, easily expanded to optimize other traits of those proteins, or employed on another problem entirely where mutations have a notable fitness cost.

In this work, we examine Monte Carlo tree search (MCTS) as an adaptive benchmark for competitive coevolution, both as a method for characterizing individual solutions, and as a measure of global progress throughout a coevolutionary run. We find that MCTS provides a much more practical measurement of solution quality than existing methods of comparing to randomly-sampled solutions. Additionally, we introduce a class of n-player games for which MCTS can be shown to have comparable performance for different n, allowing for the characterization of how coevolution performs in games with different numbers of players. We demonstrate these techniques for n-player Nim and an n-player variant of Othello.

Given recent large investments in space-based businesses, proliferated low Earth orbit (P-LEO) satellite constellations offering broadband internet services have become an active area of research and development. The landscape of satellite communications (SATCOM) has radically changed: unknown dynamics now enter the marketplace and interactions between constellation operators are undetermined. Although there has been extensive work comparing the performance of individual P-LEO constellations and their broad economic impacts, the temporal effects and competitive dynamics of this nascent marketplace have not been modeled. To address this need, we have created a tabletop board game that simulates both the system design of P-LEO constellations and the economic dynamics between constellation operators (players within the board game). By representing market events as game mechanics and incrementally developing the game complexity, we purged the majority of trivial game strategies and edge cases while simultaneously improving playability and engagement. This process has led to the development of a robust, multi-player game that can be used to examine and test high-level P-LEO constellation development strategies while reducing the time required for intensive software development and play-testing.

The genetic blueprint for the essential functions of life is encoded in DNA, which is translated into proteins-the engines driving most of our metabolic processes. Recent advancements in genome sequencing have unveiled a vast diversity of protein families, but compared with the massive search space of all possible amino acid sequences, the set of known functional families is minimal. One could say nature has a limited protein "vocabulary." A major question for computational biologists, therefore, is whether this vocabulary can be expanded to include useful proteins that went extinct long ago or have never evolved (yet). By merging evolutionary algorithms, machine learning, and bioinformatics, we can develop highly customized "designer proteins." We dub the new subfield of computational evolution, which employs evolutionary algorithms with DNA string representations, biologically accurate molecular evolution, and bioinformatics-informed fitness functions, Evolutionary Algorithms Simulating Molecular Evolution.

As computing power grows, the automated specialization and design of evolutionary algorithms (EAs) to tune their performance to individual problem classes becomes more attractive. To this end, a significant amount of research has been conducted in recent decades, utilizing a wide range of techniques. However, few techniques have been devised which automate the design of the overall structure of an EA. Most EA implementations rely solely on the traditional evolutionary cycle of parent selection, reproduction, and survival selection, and those with unique structures typically replace this with another static, hand-made cycle. Existing techniques for modifying the evolutionary structure use representations which are either loosely structured and highly stochastic, or which are constrained and unable to easily evolve complicated pathways. The ability to easily evolve complex evolutionary pathways would greatly expand the heuristic space which can be explored during specialization, potentially allowing for the representation of EAs which outperform the traditional cycle. This work proposes a methodology for the automated design of the evolutionary process by introducing a novel directed-graph-based representation, created to be mutable and flexible, permitting a black-box designer to produce reusable, high-performance EAs. Experiments show that our methodology can produce high-performance EAs demonstrating intelligible strategies.

Network segmentation has a variety of applications, including computer network security. A well segmented computer network is less likely to result in information leaks and more resilient to adversarial traversal. Conventionally network segmentation approaches rely on graph partitioning algorithms. However, general-purpose graph partitioning solutions are just that, general purpose. These approaches do not exploit specific topological characteristics present in certain classes of networks. Tailored partition methods can be developed to target specific domains, but this process can be time consuming and difficult. This work builds on previous research employing generative hyper-heuristics in the form of genetic programming for automating the development of customized graph partitioning heuristics by incorporating a dynamic approach to controlling the granularity of the heuristic search. The potential of this approach is demonstrated using two real-world complex network applications. Results show that the automated design process is capable of fine tuning graph partitioning heuristics that sacrifice generality for improved performance on targeted networks.

As large scale enterprise computer networks become more ubiquitous, finding the appropriate balance between user convenience and user access control is an increasingly challenging proposition. Suboptimal partitioning of users' access and available services contributes to the vulnerability of enterprise networks. Previous edge-cut partitioning methods unduly restrict users' access to network resources. This paper introduces a novel method of network partitioning superior to the current state-of-the-art which minimizes user impact by providing alternate avenues for access that reduce vulnerability. Networks are modeled as bipartite authentication access graphs and a multi-objective evolutionary algorithm is used to simultaneously minimize the size of large connected components while minimizing overall restrictions on network users. Results are presented on a real world data set that demonstrates the effectiveness of the introduced method compared to previous naive methods.

Discovering and exploiting the linkage between genes during evolutionary search allows the Linkage Tree Genetic Algorithm (LTGA) to maximize crossover effectiveness, greatly reducing both population size and total number of evaluations required to reach success on decomposable problems. This paper presents a comparative analysis of the most prominent LTGA variants and a newly introduced variant. While the deceptive trap problem (Trap-k) is one of the canonical benchmarks for testing LTGA, when LTGA is combined with applying steepest ascent hill climbing to the initial population, as is done in all significant LTGA variations, trap-k is trivially solved. This paper introduces the deceptive step trap problem (StepTrap-k,s), which shows the novel combination of smallest first subtree ordering with global mixing (LTS-GOMEA) is effective for black box optimization, while least linked first subtree ordering (LT-GOMEA) is effective on problems where partial reevaluation is possible. Finally, nearest neighbor NK landscapes show that global mixing is not effective on problems with complex overlapping linkage structure that cannot be modeled correctly by a linkage tree, emphasizing the need to extend how LTGA stores linkage to allow the power of global mixing to be applied to these types of problems.

For a given program, testing, locating the errors identified, and correcting those errors is a critical, yet expensive process. The field of Search Based Software Engineering (SBSE) addresses these phases by formulating them as search problems. The Coevolutionary Automated Software Correction (CASC) system targets the correction phase by coevolving test cases and programs at the source code level. This paper presents the latest version of the CASC system featuring multi-objective optimization and an enhanced representation language. Results are presented demonstrating CASC's ability to successfully correct five seeded bugs in two non-trivial programs from the Siemens test suite. Additionally, evidence is provided substantiating the hypothesis that multi-objective optimization is beneficial to SBSE.

Mate selection is a key step in evolutionary algorithms which traditionally has been panmictic and based solely on fitness. Various mate selection techniques have been published which show improved performance due to the introduction of mate restrictions or the use of genotypic/phenotypic features. Those techniques typically suffer from two major shortcomings: (1) they are fixed for the entire evolutionary run, which is suboptimal because problem specific mate selection may be expected to outperform general purpose mate selection and because the best mate selection configuration may be dependent on the state of the evolutionary run, and (2) they require problem specific tuning in order to obtain good performance, which often is a time consuming manual process. This paper introduces two versions of Learning Individual Mating Preferences (LIMP), a novel mate selection technique in which characteristics of good mates are learned during the evolutionary process. Centralized LIMP (C-LIMP) learns at the population level, while Decentralized LIMP (D-LIMP) learns at the individual level. Results are presented showing D-LIMP to outperform a traditional genetic algorithm (TGA), the Variable Dissortative Mating Genetic Algorithm (VDMGA), and C-LIMP on the DTRAP and MAXSAT benchmark problems, while both LIMP techniques perform comparably to VDMGA on NK Landscapes.

Traditional evolutionary algorithms are powerful problem solvers that have several fixed parameters which require prior specification. Determining good values for any of these parameters can be difficult, as these parameters are generally very sensitive, requiring expert knowledge to set optimally without extensive use of trial and error. Parameter control is a promising approach to achieving this automation and has the added potential of increasing EA performance based on both theoretical and empirical evidence that the optimal values of EA strategy parameters change during the course of executing an evolutionary run. While many methods of parameter control have been published that focus on removing the population size parameter, μ, all hampered by a variety of problems. This paper investigates the benefits of making μ a dynamic parameter and introduces two novel methods for population control. These methods are then compared to state-of-the-art population sizing EAs, exploring the strengths and weaknesses of each.

This paper presents the Coevolutionary Automated Software Correction system, which addresses in an integral and fully automated manner the complete cycle of software artifact testing, error location, and correction phases. It employs a coevolutionary approach where software artifacts and test cases are evolved in tandem. The test cases evolve to better find flaws in the software artifacts and the software artifacts evolve to better behave to specification when exposed to the test cases, thus causing an evolutionary arms race. Experimental results are presented on the same test problem employed in the published results on the previous state-of-the-art automated software correction system.

Parameter control in evolutionary algorithms (EAs) has been shown to be beneficial; however, the control of offspring size has so far received very little attention. This paper introduces Futility-Based Offspring Sizing (FuBOS), a method for controlling offspring size on a per generation basis without even requiring the user to set an initial offspring size value. We show that on several complex problems, an EA employing FuBOS performs on par with a highly tuned, fixed offspring size EA while being far more efficient in terms of fitness evaluations and much easier to tune as well.

Recent work in test based coevolution has focused on employing ideas from multi-objective optimization in coevolutionary domains. So called Pareto coevolution treats the coevolving set of test cases as objectives to be optimized in the sense of multi-objective optimization. Pareto coevolution can be seen as a relaxation of traditional multi-objective evolutionary optimization. Rather than being forced to determine the outcome of a particular individual on every objective, pareto coevolution allows the examination of an individual's outcome on a particular objective.

This poster paper presents a methodology for removing user-defined parameters at the parent selection stage, by allowing all individuals in the population to self-organize into pairs of mates.

The No-Free-Lunch theorem is a fundamental result in the field of black-box function optimization. Recent work has shown that coevolution can exhibit free lunches. The question as to which classes of coevolution exhibit free lunches is still open. In this paper we present a novel framework for analyzing No-Free-Lunch like results for classes of coevolutionary algorithms. Our framework has the advantage of analyzing No-Free-Lunch like inquiries in terms of solution concepts and isomorphisms on the weak preference relation on solution configurations. This allows coevolutionary algorithms to be naturally classified by the type of solution they seek. Using the weak preference relation also permits us to present a simpler definition of performance metrics than that used in previous coevolutionary No-Free-Lunch work, more akin to the definition used in the original No-Free-Lunch theorem.

While coevolution has many parallels to natural evolution, methods other than those based on evolutionary principles may be used in the interactive fitness setting. In this paper we present a generalization of coevolution to co-optimization which allows arbitrary black-box function optimization techniques to be used in a coevolutionary like manner.

One of the obstacles to Evolutionary Algorithms (EAs) fulfilling their promise as easy to use general-purpose problem solvers, is the difficulty of correctly configuring them for specific problems such as to obtain satisfactory performance. Having a mechanism for automatically configuring parameters and operators of every stage of the evolutionary life-cycle would give EAs a more widely spread popularity in the non-expert community. This paper investigates automatic configuration of one of the stages of the evolutionary life-cycle, the parent selection, via a new concept of semi-autonomous parent selection, where mate selection operators are encoded and evolved as in Genetic Programming. We compare the performance of the EA with semi-autonomous parent selection to that of a manually configured EA on three common test problems to determine the "price" we pay for user-friendliness.

The number of parameters that need to be manually tuned to achieve good performance of evolutionary algorithms and the dependency of the parameters on each other make this potentially robust and efficient computational method very time consuming and difficult to use. This paper introduces a greedy population sizing method for evolutionary algorithms (GPS-EA), an automated population size tuning method that does not require any population size related parameters to be specified or manually tuned a priori. Theoretical analysis of the number of function evaluations needed by the GPS-EA to produce good solutions is provided. We also perform an empirical comparison of the performance of the GPS-EA to the performance of an EA with a manually tuned fixed population size. Both theoretical and empirical results show that using GPS-EA eliminates the need for manually tuning the population size parameter, while finding good solutions. This comes at the price of using twice as many function evaluations as needed by the EA with an optimal fixed population size; this, in practice, is a low price considering the amount of time and effort it takes to find this optimal population size manually.

The world is increasingly dependent on critical infrastructures such as the electric power grid, water, gas, and oil transport systems, which are susceptible to cascading failures that can result from a few faults. Due to the combinatorial complexity in the search spaces involved, most traditional search techniques are inappropriate for identifying these faults and potential protections against them. This paper provides a computational methodology employing competitive coevolution to simultaneously identify low-effort, high-impact faults and corresponding means of hardening infrastructures against them. A power system case study provides empirical evidence that our proposed methodology is capable of identifying cost effective modifications to substantially improve the fault tolerance of critical infrastructures.

There exist a number of high-performance Multi-Objective Evolutionary Algorithms (MOEAs) for solving Multi-Objective Optimization (MOO) problems; two of the best are NSGA-II and epsilon-MOEA. However, they lack an archive population sorted into levels of non-domination, making them unsuitable for construction problems where some type of backtracking to earlier intermediate solutions is required. In this paper we introduce our Stored Non-Domination Level (SNDL) MOEA for solving such construction problems. SNDL-MOEA combines some of the best features of NSGA-II and epsilon-MOEA with the ability to store and recall intermediate solutions necessary for construction problems. We present results for applying SNDL-MOEA to the Tight Single Change Covering Design (TSCCD) construction problem, demonstrating its applicability. Furthermore, we show with a detailed performance comparison between SNDL-MOEA, NSGA-II, and epsilon-MOEA on two standard test series that SNDL-MOEA is capable of outperforming NSGA-II and is competitive with epsilon-MOEA.