Optimized Variational Autoencoder–Reinforcement Learning Architectures for Autonomous IoT Resource Allocation and Load Balancing

The Internet of Things (IoT) is an increasingly rapidly expanding space with a broad variety of heterogeneous devices, including the resource-constrained edge sensors and the moderately capable edge servers linked by distributed networks. To enable autonomous and decentralized allocation of resources and load balancing in IoT, this chapter proposes a novel framework that combines variational autoencoders (VAEs) with deep reinforcement learning (DRL). Increased efficiency in terms of energy consumption was a major practical benefit. The suggested VAE-DRL model was corroborated to have significantly enhanced improvement in resource utilization efficiency, energy consumption and service latency compared with traditional heuristic techniques and resource-comparative machine learning techniques. Finally, the reward function weights were manually configured, and adaptive or learning-based weighting strategies remain an open direction for future research to improve flexibility across diverse deployment settings.