Exploring the applications of deep reinforcement learning and quantum variational circuit In quantum machine learning

Abstract

In recent years, quantum computing has outperformed classical computing in many aspects, including the advancement of approaches in Reinforcement Learning prob- lems. Particularly, it has the power to utilize the quantum phenomena of super- position and entanglement, that can fastened the calculation of a vast amount of data which is very challenging for classical computers. Unfortunately, the current Quantum Computing platforms are very complex to initiate classical reinforcement learning problems for uncontrollability and intricacy of quantum circuits. In our work, we explore the application of Quantum Variational Circuit (QVC) in Deep Q- Network (DQN) instead of classical Reinforcement Learning approaches to enhance the performance of Reinforcement Learning. To achieve that, we use Quantum Vari- ational Circuit (QVC) based reinforcement learning approaches to solve the classical problems and we also solve the classical problems using classical DQN and Double Deep Q-Network (DDQN) Reinforcement Learning to compare between classical and quantum approaches. We solve Atari and Lunar Lander in OpenAI Gym envi- ronments using QVC based DQN Reinforcement learning. We study encoding tech- niques such as amplitude encoding, scaled encoding and directional encoding which were previously used in this paper[1]. We exercise IBM's open-source SDK (QISKit) and IBM-Q for quantum circuit implementation which can produce improved appli- cations like Quantum error Correction codes etc. We also use TensorFlow Quantum to implement the hybrid classical-quantum computation and experimentally analyze our work.