Global Journal of Computer Science and Technology, D: Neural & Artificial Intelligence, Volume 22 Issue 1

Global Journal of Computer Science and Technology Volume XXII Issue I Version I 31 ( )D Year 2022 © 2022 Global Journals Neural Network Design using a Virtual Reality Platform Luigi Bibbò α & Francesco Carlo Morabito σ Abstract- The evolution of Deep Learning (DL), a subset of machine learning, has made their use very effective in many artificial intelligence (AI) fields. In parallel Virtual Reality is going wide in many applications thanks to the proliferation of cameras in mobile devices and improved processing efficiency. Data visualization in deep learning is a fundamental element for which it can benefit from the advantages offered by the visualization of the VR for the development of the models. In addition, the researchers can widely use the editing of images and videos in the machine learning process to design a convolutional network suitable for image recognition. In this study, we want to demonstrate the usefulness of this approach in collecting data within virtual reality to train and optimize a convolutional neural network used to recognize human activities (HAR). Keywords: machine learning, deep learning, neural nets, visualization, virtual reality. I. I ntroduction achine Learning (ML) indicates a research area within Artificial Intelligence needs to create systems capable of autonomous learning without being specifically programmed to carry out this task [1]. This step introduces a new programming vision, in which the goal is no longer to feed input to receive an output from the machine but to allow it to make decisions autonomously. Deep Learning (DL) is a machine learning technique used to build Artificial Intelligence (AI) systems [2]. It is a technology that creates learning models on several levels. Deep learning systems allow a representation of information at various levels in a hierarchical way and build models that exploit large amounts of visual media data. The learning of data takes place through the use of statistical calculation algorithms. The inspiring principle was to reproduce a system of reasoning that is biologically inspired by the human brain, must be able to behave by simulating the functioning of neurons. The human neuron represents the computational engine that is the basis of deep learning that takes place in Artificial Neural Networks (ANN). A neural network reproduces all those processes in the brain during the learning phase and subsequent recognition phase. We design the artificial neural networks (ANN) to perform complex analysis of large amounts of data by passing it through multiple layers of neurons. The peculiar characteristic of neural networks is learning and adapting to their environment. Another property is to build connectivity models based on the Author α : DIIES University “MEDITERRANEA” Reggio Calabria, Italy. e-mail: luigi.bibbo@unirc.it Author σ: DICEAM University “MEDITERRANEA” Reggio Calabria, Italy. approximation of the error. These elements make identifying similar patterns in the test data [3]. A neural network requires a training phase and an inference phase to operate correctly. During the training phase, the number of neurons and the levels that will comprise the neural network are defined and exposed to the labeled training data. Finally, the network will learn which data is correct and discard the wrong one. Once the network has been trained, it is possible to proceed to the inference phase in which the network itself is called to evaluate new images proposed to it. These steps are part of the neural network design process, which includes: 1) Determination of the function that the network must be able to perform : a) Classification , this function involves sorting images into various classes and grouping them according to common properties, allowing you to determine whether a specific type of object is present in an image or not; b) Detection and localization , this function allows you to identify the position and size of an object by identifying the characteristics of the image; c) Segmentation, used to identify which pixels of an image belong to the corresponding objects, allows determining the boundaries and areas of the objects in the images. 2) Choice of framework: we must make this choice based on the applications to be created, the libraries available, or the developers' competence. In addition, there are numerous accessible frameworks such as, e.g., TensorFlow of Google, Caffe2 of Facebook e Open Vino of Intel, and Pytorch: an open-source solution that is now part of Facebook. 3) Loading training data : this phase and the training are crucial for the proper functioning of the neural networks. While being favored by the ability to generate features as they learn automatically, deep networks still require large amounts of training data to develop a properly functioning model. The amount of data needed depends on the complexity of the domain we are trying to approximate. The choice of the size of the dataset is closely related to the selection of the number of neurons in the neural network. The total complexity is generally not known before starting the training. For this reason, the M