Gradient-based optimization for quantum architecture search.

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Algorithm Design

Background:

• Quantum Architecture Search (QAS) is crucial for optimizing quantum circuits in Variational Quantum Algorithms (VQAs).
• Current QAS methods are inefficient due to their exploration of discrete circuit spaces.
• A need exists for more efficient and scalable QAS techniques.

Purpose of the Study:

• To propose Gradient-based Optimization for Quantum Architecture Search (GQAS), a novel approach for quantum circuit design.
• To enable efficient exploration of quantum circuit architectures in a continuous latent space.
• To improve the performance of Variational Quantum Algorithms through optimized circuit design.

Main Methods:

• Developed a GQAS framework utilizing a circuit encoder, decoder, and predictor.
• Embedded quantum architectures into a continuous latent space using an encoder.
• Optimized latent representations via gradient descent and mapped back to discrete architectures using a decoder.
• Pre-trained the encoder using Self-Supervised Learning (SSL) on a large dataset of circuit architectures.

Main Results:

• GQAS demonstrated superior performance compared to existing Differentiable Quantum Architecture Search (DQAS) methods.
• Simulations on the Variational Quantum Eigensolver (VQE) validated the effectiveness of the proposed approach.
• The continuous latent space optimization proved more efficient than discrete search methods.

Conclusions:

• GQAS offers a more efficient and effective method for designing quantum circuits for VQAs.
• The proposed gradient-based approach in a continuous latent space represents a significant advancement in QAS.
• This work paves the way for designing more complex and higher-performing quantum circuits.