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

Neural Network Design using a Virtual Reality Platform Global Journal of Computer Science and Technology Volume XXII Issue I Version I 34 ( )D Year 2022 To attribution methods, Chattopadhyay and al. [19] have developed a methodology that starts from the visualization of the neural network's architecture as a Structural Causal Model (SCM) and calculates each feature's causal effect on the output. Their approach differs from other related works to structural learning [20]. The goal is to discern the causal structure in data provided; their goal is to identify the causal influence on the output of a learned function. Sattarzadeh and al. [21] propose a model based on visualization maps from multiple levels using an attribution-based input sampling technique and aggregating them to achieve a satisfactory and complete explanation. The methodology adopted to interpret the layers of CNN results in a four-step method of description. In the first three stages, information extracted from multiple levels of CNN is represented in the accompanying display maps. These maps are then merged into a single module to form an explanation map in the last step. The proposed solution allows overcoming some problems encountered in models that use input sampling techniques. These methods have shown excellent fidelity in rationally inferring model predictions. However, they exhibit instability indices since their output depends on sampling randomness (RISE) or random initialization to optimize a perturbation mask. Finally, their algorithm called Semantic Input Sampling for Explanation (SISI) replaces the randomized RISE input sampling technique with an attribution-based sampling technique. Finally, it uses feature maps derived from multiple levels. The Deep Replay open-source Python package was designed to allow you to visualize how a Deep Learning model in Keras is performing or learning at each iteration/epoch. Virtual reality (VR), thanks to its peculiar characteristics, represents a valid alternative to the visualization of conventional data. VR technology establishes a new form of human-computer interaction, arousing in the user a new type of experience linked to the concept of "presence” [22]. Presence is the sensation of being physically and spatially placed in an environment. With virtual reality, you can realize this feeling of being in another world, the virtual one. VR can generate experience and support knowledge acquisition thanks to this new communication process. However, the sense of presence developed by VR depends significantly on the characteristics of the technology used. It is associated with the perceptual illusion that the user has to interact with the remote environment, such as if it were present. In the interaction with the VR environment, a perceptual illusion is created in the user. The stimulation of the senses produces cognitive and emotional models consistent with the experimenting environment. When in VR, a user will inevitably compare the appearance of virtual objects with real-world objects and judge the level of congruence. The other element that favors VR over other visualization techniques is the level of "immersion"[23]. The greater immersion allows you to improve the design business, perform tasks more efficiently and with greater understanding. Immersion in virtual reality is the perception of being physically present in a non-physical world. Perception is created by surrounding the user of the VR system with images, sounds, or other stimuli that provide an evocative environment. Finally, the user can move freely and explore places and spaces through the movements of the head and eyes and interact with objects by grabbing and dragging them. In addition, movements from the physical world are transferred to the virtual world with great precision. This aspect is precisely exploited to simulate physical actions. Finally, workflows and activities can be transferred to the virtual model to verify the correct interpretation, improving the system's accuracy. These characteristics make virtual reality a suitable environment for simulations and e-learning purposes as they can stimulate users' creativity and accelerate the learning process [24]. This technology is widely applied in various fields ranging from entertainment design to education and rehabilitation and psychological therapies [25]. VR technology makes it possible to improve the rehabilitation-psychological program compared to traditional techniques by creating a more natural process, reproducing the characteristics of home living environments. Other elements that characterize this technique are represented by the possibility of dynamically adjusting the difficulty of the exercises concerning the skills acquired and stimulating the patient's multisensory skills. In the context of training, it has proved to be very effective compared to traditional techniques as they can keep the attention high and focus the topics better [26]. Immersion and interactivity are the elements that suggest the choice of VR for CNN as through them; the user can better focus and understand how it works. The following sections illustrate the main characteristics of CNN and virtual reality technologies. III. C onvolutional N eural N etwork From an architectural point of view, a convolutional neural network is a multi-layered feedforward neural network composed of an input layer, hidden layers, and an output layer. The hidden layers are typically convolutional, followed by activation and pooling layers. This sequential design allows convolutional neural networks to learn hierarchical features. Convolutional neural networks (CNN) process data through many layers of artificial neurons. This human brain (CNN) process is constituted by a set of different layers that act as extractors of the features of the input images and a fully connected terminal network that acts as a classifier. It has proved to be an effective © 2022 Global Journals