A Cache-Enabled Device-to-Device Approach Based on Deep Learning

Abstract

In this paper, we present a deep learning-based Device-to-Device (D2D) approach that utilizes Gated Recurrent Unit (GRU) model that is optimized through Bayesian optimization for hyperparameter tuning. The proposed approach, DLCE-D2D (Deep Learning Cache-Enabled device-to-device) system utilizes deep learning using GRU, to predict the popularity of content in a D2D network and dynamically adjusts the cache eviction policy to improve the cache hit ratio. DLCE-D2D approach was evaluated using real-world data and compared against traditional cache eviction policies such as Least Recently Used (LRU) and First In First Out (FIFO). The results show that the proposed approach outperforms traditional policies in terms of cache hit ratio. Also, it was demonstrated that DLCE-D2D approach is robust against changes in data access patterns and can adapt to dynamic changes in the network. Additionally, it was shown that the GRU model can achieve similar or better results than other deep learning-based methods. The use of Bayesian optimization for hyperparameter tuning in the proposed approach offers a promising solution for improving the cache hit ratio in D2D networks, thereby improving the performance and reducing the cost of such networks.

Keywords

• Bayesian optimization
• deep learning
• device-to-device (D2D)
• edge network
• gated recurrent unit (GRU)