Huaman Allccahuaman, Ruth M.

Publication Search Results

Now showing 1 - 1 of 1

Data caching with deep reinforcement learning
(2024-07-09) Huaman Allccahuaman, Ruth M.; Rodríguez Martínez, Manuel; College of Engineering; Arzuaga, Emmanuel; Rodríguez, Domingo; Department of Electrical and Computer Engineering; Barriga Burgos, Alicia
The growth of data traffic in telecommunications networks is becoming an issue, pos- ing new challenges to network architectures. Cache memory is a fixed-size, high-speed storage space that stores a subset of data. Cache memory has a policy that makes the decision of what data is important to cache, this helps to have faster access to data re- quested by users. There are caching policies such as least recently used (LRU), and first in, first out (FIFO), but these do not take into account certain patterns that data has. Knowing the need to improve cache organization, this research developed a Reinforce- ment Learning model, we used the Proximal Policy Optimization (PPO) architecture with deep neural networks, within the environment we used the Long Short-Term Mem- ory (LSTM) and embeddig components. The Reinforcement Learning (RL) model was implemented and trained with different observations and rewards. The data used for training and testing are based on the Zipf distribution. Experimental test results show that our proposed model can make smarter decisions in organizing cached data. It improves hit rate based performance and long-term stability compared to LRU, FIFO and RL model policies.