Neuroevolution in Deep Neural Networks: Current Trends and Future Challenges

A variety of methods have been applied to the architectural configuration and learning or training of artificial deep neural networks (DNN). These methods play a crucial role in the success or failure of the DNN for most problems and applications. Evolutionary algorithms (EAs) are gaining momentum as a computationally feasible method for the automated optimization of DNNs. Neuroevolution is a term, which describes these processes of automated configuration and training of DNNs using EAs. While many works exist in the literature, no comprehensive surveys currently exist focusing exclusively on the strengths and limitations of using neuroevolution approaches in DNNs. Absence of such surveys can lead to a disjointed and fragmented field preventing DNNs researchers potentially adopting neuroevolutionary methods in their own research, resulting in lost opportunities for wider application within real-world deep learning problems. This article presents a comprehensive survey, discussion, and evaluation of the state-of-the-art in using EAs for architectural configuration and training of DNNs. This article highlights the most pertinent current issues and challenges in neuroevolution and identifies multiple promising future research directions.

Impact Statement—The concept of deep learning originated from the study of artificial neural networks (ANNs). ANNs have achieved extraordinary results in a variety of diverse application areas. Numerous methods have been applied to the architectural configuration and learning or training of artificial DNN and these methods play a crucial role in the success or failure of the DNN for most problems and applications. Recently, EAs have been gaining momentum as a computationally feasible method (called neuroevolution) for the automated configuration and learning or training of DNNs. This article reviews over 170 recent scientific papers describing how major EAs paradigms are being applied by researchers to the configuration and optimization of multiple DNNs. By articulating a clear understanding of the context, state-of-the-art, and feasibility of Neuroevolution, researchers in AI, EAs, and DNN will benefit from this article. The impact of this article comes from contributing toward enhancing research capacity, knowledge, and skills for researchers currently working in neuroevolution and actively engaging those considering becoming involved in this area.

I. Introduction

Deep learning (DL) algorithms [57], [65], [94], a subset of machine learning algorithms, are inspired by deep hierarchical structures of human perception as well as production systems. These algorithms have achieved extraordinary results in diverse areas including computer vision [159], speech recognition [58], [115], board games [145], and video games [114], to mention a few. The design of deep neural networks (DNNs) architectures (along with the optimization of their hyperparameters) and their training plays a crucial part in their success or failure [105]. Architecture search is an area of growing interest as demonstrated by the large number of scientific works published in recent years. These works can be classified into one of the following two broad categories: evolution-based methods [6], [34], sometimes referred as neuroevolution [42], [170], and reinforcement learning (RL) methods [158]. Methods falling outside these two categories have also been proposed in the specialized literature including Monte Carlo based simulations [119], random search [11] and random search with weight prediction [14], hill-climbing [37], grid search [174], Bayesian optimization [12], [76], gradient-based [103], [168], and mutual information [161], [162], [173]. RL architecture-search methods started gaining momentum thanks to their impressive results [7], [16], [101], [179], [181], [182], and more recently, EA architecture-search methods began yielding impressive results in the automatic configuration of DNNs architectures [39], [102], [150]. It has been reported that neuroevolution requires less computational time compared to RL methods [114], [130], [150], [155]. Basically, a DNN is a feedforward artificial neural network (ANN) with many hidden layers with each layer constituting a nonlinear information processing unit. Usually having two or more hidden layers in an ANN signifies a DNN. By adding more layers and more units within a layer a DNN can represent functions of increasing complexity [57].