Multi-Objective Evolutionary Architecture Search for Parameterized Quantum Circuits.

Area of Science:

• Quantum Computing
• Machine Learning
• Artificial Intelligence

Background:

• Parameterized quantum circuits (PQCs) show promise in hybrid quantum-classical machine learning for reinforcement learning (RL).
• However, PQC architecture design and inter-circuit interactions remain underexplored areas.
• Classical systems like deep neural networks face challenges in complex RL tasks where PQCs offer potential advantages.

Purpose of the Study:

• To develop an efficient framework for searching optimal PQC architectures for RL.
• To explore the impact of PQC architecture design on learning performance and other objectives.
• To identify critical design choices in hybrid quantum-classical learning systems.

Main Methods:

• Introduced a Multi-objective Evolutionary Architecture Search framework for Parameterized Quantum Circuits (MEAS-PQC).
• Employed a multi-objective genetic algorithm with quantum-specific configurations for efficient architecture searching.
• Evaluated the framework on three benchmark RL tasks.

Main Results:

• MEAS-PQC successfully identified PQC architectures with superior learning performance on benchmark RL tasks.
• The framework optimized architectures for reduced quantum noise and smaller model size.
• Analysis revealed performance-critical design patterns and probability distributions of quantum operations.

Conclusions:

• The proposed MEAS-PQC framework enables efficient discovery of high-performing PQC architectures for RL.
• Optimizing PQC architecture design can lead to enhanced learning, reduced noise, and smaller models.
• Understanding quantum operation patterns is key to advancing hybrid quantum-classical learning systems.