### Overview This component contains the pre-processed results of the execution of an evolutionary algorithm (EA) in a byzantine environment as described in ref. 3, in which the fitness information is unreliable. The source code can be found in [the project's GitHub repository][1]. ### Contents The folder `panmictic` contains the results for an EA run on three different configurations: - `none`: a basic EA that does not include any mechanism to handle fitness unreliability. - `majority3`: an EA using a handler based on the majority result of 3 fitness evaluations. - `average3`: an EA using a handler based on the average result of 3 fitness evaluations. Each compressed archive contains a hierarchy of folders with the structure `model`/`problem`/`p`/, where: - `model` can take the values `inverter` (malicious model in which unreliable fitness is negatively correlated with true fitness) and `randomizer` (where unreliable fitness is uncorrelated with true fitness). - `problem` is the target problem, namely `onemax` (100-bit counting-ones function), `leadingones` (100-bit Rudolph's Leading-Ones function), `trap` (25 x Deb’s 4-bit fully deceptive function), and `mmdp` (17 x Goldberg et al.’s Massively Multimodal Deceptive Problem). - `p` is the probability of a fitness evaluation returning an unreliable value. It can take values from 0% to 50% in steps of 5%. For each of these combinations, 50 runs are performed. The results are stored in separate CVS files named `runX.csv` (where `X` goes from 1 to 50), and contain the best fitness and the population entropy measured each 1000 fitness evaluations. At the root of the folder hierarchy, the file `configuration.json` contains the general EA configuration. ### Bibliography 1. C. Cotta, "[On the Performance of Evolutionary Algorithms with Unreliable Fitness Information][2]", EvoStar 2023 Late Breaking Abstracts, A.M. Mora (ed.), Brno, Czech Republic, 2023 2. C. Cotta, "Tackling Adversarial Faults in Panmictic Evolutionary Algorithms", GECCO '23 Companion: Genetic and Evolutionary Computation Conference Companion Proceedings, ACM, New York, NY, USA, 2 pages. [doi:10.1145/3583133.3596426][3] 3. C. Cotta, "[Enhancing Evolutionary Optimization Performance under Byzantine Fault Conditions][4]", 18th International Conference on Hybrid Artificial Intelligence Systems - HAIS 2023, Lecture Notes in Computer Science 14001, Springer, 2023 (forthcoming) [1]: https://github.com/Bio4Res/ea-model-byzantine [2]: http://www.lcc.uma.es/~ccottap/papers/cotta23performance.pdf [3]: http://doi.org/10.1145/3583133.3596426 [4]: http://www.lcc.uma.es/~ccottap/papers/cotta23enhancing.pdf