Bi-level Multi-objective Evolution of a Multi-layered Echo-State Network Autoencoder for Data Representations

•ML-ESN is used as an Autoencoder (ML-ESNAE) to extract new data features.•A bi-level evolutionary optimization based on PSO is proposed for architecture and parameters optimization.•Level 1 offers a multi-objective architecture optimization, providing maximum accuracy while maintaining a reduced complexity.•Every Pareto optimal solution obtained from level 1 undergoes a mono-objective weights optimization at level 2.•An empirical study shows that the evolved ML-ESNAE produces an improvement in extracting more expressive features.