الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 169 |  |  |
| | | from | 169 | | |
Abstract
This thesis addresses the problem of improving caching networks through the application of machine learning and deep learning techniques. Caching networks play a vital role in content delivery systems, and optimizing their performance is crucial for efficient data access and reduced network congestion. The primary challenge lies in determining where to cache content, what content to cache, and how to make caching decisions dynamically. Traditional approaches have limitations in adapting to changing network conditions and user demands. To overcome these challenges, this research proposes a novel solution based on deep learning algorithms.
This paper introduces a cache-enabled Device-to-Device (D2D) approach based on an optimized cascaded Gated Recurrent Unit (GRU) model for content caching in edge networks. The proposed approach utilizes a cascaded GRU model to accurately predict content popularity and optimize cache placement. Bayesian optimization is employed to enhance the performance of the GRU model by optimizing its hyperparameters.
The evaluation of the approach using the MovieLens dataset, which consists of 100,000 ratings of 9,000 movies by 600 users, reveals its effectiveness. The Movie Rank Category Prediction predictor achieves an impressive test accuracy of 98.5% with an F1-score of 0.97, precision of 0.977, and recall of 0.962, accurately classifying movie rank categories. The Request Count predictor estimates the number of requests for a movie with reasonable accuracy, achieving an RMSE of 1.17. The cascaded predictor for caching decisions, utilizing GRU models, outperforms CNN and LSTM models with an RMSE of 1.2.
The proposed model with D2D feature enabled achieves a cache hit rate of 73% using only 25% of the cache capacity, demonstrating efficient utilization of limited cache resources. Enabling D2D further improves the cache hit rate by 5%. Additionally, the model with D2D feature outperforms the D2D-disabled approach after deploying 10% of the cache size compared to the total data size. Overall, the results highlight the superior performance of the DLCE-D2D approach in optimizing cache utilization and improving the cache hit ratio in edge networks.
In conclusion, this thesis presents a comprehensive study on leveraging deep learning techniques to enhance caching networks. The proposed solution offers an effective approach for dynamically optimizing caching decisions and improving content delivery performance. The quantitative results validate the effectiveness and superiority of the proposed solution, emphasizing its potential for practical implementations in caching networks.