Neuroevolution in Deep Neural Networks: Current Trends and Future Challenges

Abstract

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.

InfoNEAT: Information Theory-based NeuroEvolution of Augmenting Topologies for Side-channel Analysis

“…In that sense, the architecture of the network (i.e., the number of layers and number of hidden nodes) can also be considered as part of the search space since they affect the size of the network. Neuroevolution techniques such as NEAT evolve both the weights and NN architecture through crossover and mutation [20]. Thus, compared to traditional optimizers (such as the stochastic gradient descent, SGD) which only optimize the weight, the search space for neuroevolution techniques is considerably larger [1].…”

“…This is accomplished by assigning a fitness value to each genome. Afterward, the fitness values are taken into account by a selection operator, e.g., tournament selection [20] (see Figure 3(c)). Crossover (lines #6 in Algorithm 1): The steps corresponding to fitness evaluation and selection are then followed by the crossover operation, also known as recombination, where the genetic information (e.g., the parameters of the NNs) of two selected individuals are combined as shown in Figure 3(d).…”

DECORE: Deep Compression with Reinforcement Learning

“…Figure 4 shows the number of channels in the compressed and pre-trained VGG network on CIFAR10. We found that for deeper layers (11)(12)(13)(14)(15) the number of channels in the compressed network is <1% of the pre-trained network. This shows that the network is not learning much in deeper layers as the input resolution (2X2) for these layers is very small and does not provide enough information to learn unique features.…”

“…Some approaches to automated architecture design have relied on sparsifying regularizers, often applying L1 regularization to obtain a sparse, subnetwork that can be extracted from the original large network [6,8,9]. Extensions to this work have also taken into account resource constraints such as latency and computation [5,10,11]. Other approaches build networks from the ground up using a set of custom building blocks, relying on variations of trial and error search to find promising architectures [12][13][14][15].…”

Direct evaluation of progression or regression of disease burden in brain metastatic disease with Deep Neuroevolution

“…Deep neuroevolution (DNE) is a promising sub-field of evolutionary strategies/genetic algorithms that improves the performance of convolutional neural networks (CNNs) [25]. CNNs, along with transformers, currently dominate deep learning applications in radiology.…”