Fidelity-based probabilistic Q-learning for control of quantum systems.

Area of Science:

• Quantum Control
• Reinforcement Learning
• Machine Learning

Background:

• The exploration-exploitation dilemma is a central challenge in reinforcement learning, particularly for Q-learning algorithms.
• Effective strategies are needed to balance exploring new actions and exploiting known ones for optimal policy learning.

Purpose of the Study:

• To introduce a novel fidelity-based probabilistic Q-learning (FPQL) approach for reinforcement learning.
• To apply FPQL for the learning control of quantum systems.
• To address the exploration-exploitation balance in reinforcement learning for quantum applications.

Main Methods:

• Developed a probabilistic Q-learning (PQL) algorithm to illustrate probabilistic action selection.
• Introduced the fidelity-based probabilistic Q-learning (FPQL) algorithm, using fidelity to guide learning.
• Iteratively updated action selection probabilities based on learning progress.

Main Results:

• FPQL demonstrated a superior balance between exploration and exploitation compared to traditional methods.
• The algorithm successfully avoided local optimal policies in quantum system control.
• FPQL accelerated the learning process for controlling quantum systems.

Conclusions:

• The FPQL approach offers a natural and effective solution to the exploration-exploitation problem in reinforcement learning.
• FPQL shows significant promise for advancing the field of quantum control through improved learning strategies.
• The method's ability to avoid local optima and speed up learning makes it a valuable tool for complex control tasks.