| openaire: EC/HE/101141916/EU//Artificial User

Neural Architecture Search (NAS) is a busy research field growing exponentially in recent years. State-of-the-art deep neural networks usually require a specialist to fine-tune the model to solve a specific problem. NAS research aims to design neural network architectures automatically, thus easing the need for machine learning specialists to spend a lot of effort on hand-crafted attempts. As artificial intelligence applications are becoming ubiquitous, there is also a growing interest in efficient applications that could be deployed to smartphones, smart wearable devices, and other edge devices. Gender recognition in unfiltered images — such as those we find in real-world situations like pictures taken with smartphones and video shots from surveillance cameras — is one of such challenging applications. In this work, we developed an evolutionary NAS algorithm that consistently finds efficient ResNet-based architectures, named RENNAS, which have a good trade-off between classification accuracy and architectural and computational complexities. We demonstrate our algorithm's performance on Adience dataset of unfiltered images for gender recognition.

Currently, Deep Convolutional Neural Networks (DCNNs) are used to solve all kinds of problems in the field of machine learning and artificial intelligence due to their learning and adaptation capabilities. However, most successful DCNN models have a high computational complexity making them difficult to deploy on mobile or embedded platforms. This problem has prompted many researchers to develop algorithms and approaches to help reduce the computational complexity of such models. One of them is called filter pruning, where convolution filters are eliminated to reduce the number of parameters and, consequently, the computational complexity of the given model. In the present work, we propose a novel algorithm to perform filter pruning by using Multi-Objective Evolution Strategy (ES) algorithm, called DeepPruningES. Our approach avoids the need for using any knowledge during the pruning procedure and helps decision-makers by returning three pruned CNN models with different trade-offs between performance and computational complexity. We show that DeepPruningES can significantly reduce a model's computational complexity by testing it on three DCNN architectures: Convolutional Neural Networks (CNNs), Residual Neural Networks (ResNets), and Densely Connected Neural Networks (DenseNets).

Deep Convolutional Neural Networks (DCNNs) have the potential to revolutionize the field of Medical Imaging Diagnostics due to their capabilities of learning by using only raw data. However, DCNNs can only learn when trained using thousands of data points, which is not always available when dealing with medical data. Moreover, due to patient privacy concerns and the small prevalence of certain diseases in the population, medical data often presents unbalanced classes and fewer data points than other data types. Researchers often rely on Generative Adversarial Networks (GANs) to synthesize more data from a given distribution to solve this problem. Nevertheless, GANs are computationally intensive models requiring the use of powerful hardware to run. In the present work, an algorithm for pruning GANs based on Evolution Strategy (ES) and Multi-Criteria Decision Making (MCDM) is proposed in which a model with the best trade-off between computational complexity and synthesis performance can be found without the use of any trade-off parameter. In the proposed algorithm, the model with the best trade-off is defined geometrically as the candidate solution with the minimum Manhattan distance (MMD) in a two-dimensional objective space established by the number of Floating-Point Operations (FLOPs) and the Wasserstein distance of all candidate solutions, also known as the knee solution. The results show that the pruned GAN model achieves similar performance compared with the original model with up to 70% fewer Floating-Point Operations.

The field of medical imaging diagnostic makes use of a modality of imaging tests, e.g., X-rays, ultrasounds, computed tomographies, and magnetic resonance imaging, to assist physicians with the diagnostic of patients' illnesses. Due to their state-of-the-art results in many challenging image classification tasks, deep neural networks (DNNs) are suitable tools for use by physicians to provide diagnostic support when dealing with medical images. To further advance the field, the present work proposes a two-phase algorithm capable of automatically generating compact DNN architectures given a database, called here DNNDeepeningPruning. In the first phase, also called the deepening phase, the algorithm grows a DNN by adding blocks of residual layers one after another until the model overfits the given data. In the second phase, called the pruning phase, the algorithm prunes the created DNN model from the first phase to produce a DNN with a small amount of floating-point operations guided by some preference given by the user. The proposed algorithm unifies the two separate fields of DNN architecture searching and pruning under a single framework, and it is tested in two medical imaging data sets with satisfactory results.