# Neuroevolution in Deep Neural Networks: Current Trends and Future Challenges

**Authors:** Edgar Galván, Peter Mooney  
**arXiv:** [2006.05415](https://arxiv.org/abs/2006.05415)  
**Date:** 2020-06-09  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2006.05415v1/

## Key points

- Authors Edgar Galván and Peter Mooney explore the application of Evolutionary Algorithms (EAs) for optimizing and training artificial deep neural networks (DNNs).
- Neuroevolution techniques play a crucial role in determining the success or failure of DNNs across various problems and applications.
- There is a lack of comprehensive surveys focusing exclusively on the strengths and limitations of using neuroevolution in optimizing DNN performance.
- Galván and Mooney present a thorough survey, discussion, and evaluation of state-of-the-art works that utilize EAs for configuring and training DNNs.
- Their analysis identifies current challenges in neuroevolution while pointing towards promising avenues for future exploration.
- The research sheds light on the evolving landscape of neuroevolution within the realm of deep neural networks, offering insights to optimize architectural design and training methodologies.
- Galván and Mooney provide valuable perspectives that can guide advancements in leveraging evolutionary algorithms to enhance the capabilities and applicability of DNNs.

## Summary

In their paper titled "Neuroevolution in Deep Neural Networks: Current Trends and Future Challenges," authors Edgar Galván and Peter Mooney explore the application of Evolutionary Algorithms (EAs) for optimizing and training artificial deep neural networks (DNNs). They emphasize the crucial role of neuroevolution techniques in determining the success or failure of DNNs across various problems and applications. While EAs are increasingly recognized as a feasible approach for enhancing DNN performance, there is a lack of comprehensive surveys focusing exclusively on the strengths and limitations of using neuroevolution in this context. This gap can hinder researchers from adopting these methods in their own work and potentially limit opportunities to improve deep learning's real-world applicability. To address this issue, Galván and Mooney present a thorough survey, discussion, and evaluation of state-of-the-art works that utilize EAs for configuring and training DNNs. By synthesizing existing research, they identify current challenges in neuroevolution while also pointing towards promising avenues for future exploration. Their analysis sheds light on the evolving landscape of neuroevolution within the realm of deep neural networks, offering insights that can inform researchers seeking to optimize their approaches to architectural design and training methodologies. Through this comprehensive examination, Galván and Mooney provide valuable perspectives that can guide advancements in leveraging evolutionary algorithms to enhance the capabilities and applicability of DNNs in diverse domains.