8000 GitHub - marcomistretta/Quantum-DQN: This repository implements a novel approach to the Cart Pole problem by integrating quantum computing principles into the Deep Q Network. Enhance exploration and policy learning for efficient balancing. Dive in and explore the intersection of quantum computing and deep reinforcement learning.
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

This repository implements a novel approach to the Cart Pole problem by integrating quantum computing principles into the Deep Q Network. Enhance exploration and policy learning for efficient balancing. Dive in and explore the intersection of quantum computing and deep reinforcement learning.

Notifications You must be signed in to change notification settings

marcomistretta/Quantum-DQN

Repository files navigation

Exploring the Potential of Quantum Computing for Reinforcement Learning: A Case Study on the Cart Pole Environment!

Deep Reinforcement Learning (RL) has recently achieved impressive results on a variety of challenging tasks. However, the computational resources required for Deep RL can be huge, especially when the RL tasks involve high dimensional observation spaces or long
time horizons. To overcome these limitations, a promising approach is to use Quantum computers to accelerate the training of DNNs for RL.

In this paper, we re-implement the quantum circuit introduced in Quantum agents in the Gym: a variational quantum algorithm for deep Q-learning to learn the Cart Pole Environment. The algorithm combines Variational Quantum Algorithms (VQAs) and Deep Q-learning (DQL). Our Quantum agent is implemented in Qiskit, an open-source quantum computing framework.

UML diagram

UML

Quantum Circuit

QuantumCircuit

Ansatz

Ansatz

Comparison Result with Classical Approach

Comparison

Comparison between Quantum (Left) and Classical (Right) Approach.

From the graphs above, it’s evident that quantum model training is globally more robust than its classical counterpart. However the classic model is undoubtedly faster than the quantum approach despite it has 17’000 trainable parameters, unlike the quantum approach which has only 46 trainable parameters.

Contributors

Supervisor

About

This repository implements a novel approach to the Cart Pole problem by integrating quantum computing principles into the Deep Q Network. Enhance exploration and policy learning for efficient balancing. Dive in and explore the intersection of quantum computing and deep reinforcement learning.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages

0