Federated Learning on Virtual Reality Environments: Performance Analysis on Standalone Devices

Federated Learning on Virtual Reality Environments: Performance Analysis on Standalone Devices

연구 분야: Analysis

학회: International Conference on Web Engineering

Training algorithms through Federated Learning has emerged as a promising strategy to safeguard data privacy in distributed environments. This training can be performed on several devices, ranging from high-capacity servers to devices with limited capabilities. However, handling numerous data sources can overload these devices, especially low-power ones, increasing response time. A particular scenario is Virtual Reality, as it requires connection to multiple data sources where latency is critical. Virtual Reality devices have traditionally required a continuous connection to computer equipment, limiting their versatility and the advantages of wireless devices. Recent technological advancements in these devices have increased their computational capabilities, enabling them to perform certain tasks independently. This work addresses the challenge of training a neural network on Virtual Reality devices through a federated system, to develop an enriched collaborative model stored and aggregated in the Cloud. The objective is to evaluate the computational costs and discern the possibilities and limitations of Virtual Reality in Artificial Intelligence.