Query Optimization using Machine Learning Techniques

Abstract

Query optimization is an important aspect in the design of relational database management systems (DBMS), aiming to find an optimal execution plan by minimizing the total execution time of queries. With this in mind, our work involves using a new paradigm such as deep reinforcement learning (Deep RL) is a sub-domain of machine learning that combines reinforcement learning (RL) and deep learning to improve query optimization approaches which is a complete NP problem. Through this task, we aim to reimplemente and adapt DRL algorithms to prove their performance. We use the Proximal Policy Optimization (PPO) algorithm as a model-Free , and the Universal Value Function Approximators (UVFA) with Hindsight Experience Replay. The tests of our modest expreinces on the IMDB dataset allowed us to observe a gradual performance by playing on the hyperparametres of PPO such as the activation function and a slight difference in favor of UVFA with HER.