Skip to content

Releases: MarkCodering/Quantum-Maze-DQN

A1 - Deep Reinforcement Learning Using Hybrid Quantum Neural Network - 程式碼壓縮檔

2022-23 Tamkang University GEP Release

29 Apr 06:31
Compare
Choose a tag to compare

Abstract

Quantum machine learning is a novel and highly immature field of study. Its promising implication for combining two powerful areas in computer science and physics, quantum machine learning, has made an interesting topic for researchers to investigate to resolve various challenging problems. However, considering the limitations of Noisy Intermediate-Scale Quantum (NISQ) devices and machine learning algorithm designs, some experimental models haven’t demonstrated strong effectiveness and eventually surpass the present performance of a full classical model. Therefore, this research would like to investigate the potential and possibility of integrating a parameterized quantum circuit with a deep neural network to solve a reinforcement learning problem. Ultimately, this research would like to provide some potential improvements that could be made within this hybrid deep learning model. Also, the study is based on the IBM Quantum Systems (IBM-Q) ecosystem to enable rapid testing and evaluation. The quantum computation experiments were tested and trained on the full-classical hardware with IBM Qiskit’s Aer simulator. Current robotics devices could incorporate our quantum deep reinforcement learning model with this solution architecture.

Issue Resolved:

V1.0.0-Pre-Alpha

26 Apr 07:58
Compare
Choose a tag to compare
V1.0.0-Pre-Alpha Pre-release
Pre-release

About this release

Quantum computation has a strong implication for advancing the current limitation of machine learning algorithms to deal with higher data dimensions or reducing the overall training parameters for a deep neural network model. Based on a gate-based quantum computer, a parameterized quantum circuit was designed to solve a model-free reinforcement learning problem with the deep-Q learning method. This research has investigated and evaluated its potential. Therefore, a novel PQC based on the latest Qiskit and PyTorch framework was designed and trained to compare with a full-classical deep neural network with and without integrated PQC. At the end of the research, the research draws its conclusion and prospects on developing deep quantum learning in solving a maze problem or other reinforcement learning problems.