Quantum Machine Learning Architecture Search via Deep Reinforcement Learning (2407.20147v1)

Abstract: The rapid advancement of quantum computing (QC) and ML has given rise to the burgeoning field of quantum machine learning (QML), aiming to capitalize on the strengths of quantum computing to propel ML forward. Despite its promise, crafting effective QML models necessitates profound expertise to strike a delicate balance between model intricacy and feasibility on Noisy Intermediate-Scale Quantum (NISQ) devices. While complex models offer robust representation capabilities, their extensive circuit depth may impede seamless execution on extant noisy quantum platforms. In this paper, we address this quandary of QML model design by employing deep reinforcement learning to explore proficient QML model architectures tailored for designated supervised learning tasks. Specifically, our methodology involves training an RL agent to devise policies that facilitate the discovery of QML models without predetermined ansatz. Furthermore, we integrate an adaptive mechanism to dynamically adjust the learning objectives, fostering continuous improvement in the agent's learning process. Through extensive numerical simulations, we illustrate the efficacy of our approach within the realm of classification tasks. Our proposed method successfully identifies VQC architectures capable of achieving high classification accuracy while minimizing gate depth. This pioneering approach not only advances the study of AI-driven quantum circuit design but also holds significant promise for enhancing performance in the NISQ era.

Quantum Machine Learning Architecture Search via Deep Reinforcement Learning

This paper presents a novel approach to Quantum Machine Learning (QML) using a deep reinforcement learning (RL) framework. The primary objective is to automate the design of Variational Quantum Circuits (VQCs) tailored for specific supervised learning tasks, without relying on pre-established physical ansatz. This methodology emerges as a solution to traditional QML model design, which demands significant expertise to navigate the constraints of Noisy Intermediate-Scale Quantum (NISQ) devices.

Methodology

The paper introduces a reinforcement learning-based framework for Quantum Architecture Search (QAS), denoted as RL-QMLAS. This involves training an RL agent using a Double Deep Q-Network (DDQN) algorithm enhanced with an adaptive search mechanism. The agent dynamically adjusts its target learning objectives, aiming to optimize quantum gate sequences on-the-fly for improved performance in classification tasks.

1. Variational Quantum Circuits (VQCs): VQCs play a crucial role in the proposed framework. These circuits are leveraged as parameterized quantum subcircuits functioning analogously to classical neural networks. The authors emphasize that the overall circuit architecture significantly impacts QML performance.
2. State Representation: An N-step DDQN is employed to represent states as matrices outlining gate configurations, reflecting the circuit's structural configuration.
3. Adaptive Search Strategy: This strategy adjusts the accuracy target and exploration rate based on the RL agent's evolving performance. When the agent consistently meets or exceeds the current objectives, targets increment slightly, challenging the agent to further refine its solutions.

Results

Through simulations on classifications tasks using datasets like scikit-learn's make_classification and make_moons, the RL-QMLAS framework demonstrated proficiency in dynamically generating quantum circuits. The performance, in terms of classification accuracy and circuit depth, was competitive with classical machine learning models, showcasing minimal resource usage (e.g., fewer quantum gates and parameters in comparison to classical models).

• The effectiveness of the RL-QMLAS is evident in its ability to optimize VQCs without prior knowledge, achieving accuracies close to traditional models like logistic regression and support vector machines.

Implications and Future Prospects

This paper advances the frontier of AI-driven quantum circuit design, presenting a scalable approach to QML model design compatible with NISQ devices. The implications are significant: QML can be more accessible, facilitating applications across diverse domains by reducing the barrier of required expertise.

Looking forward, the versatility of the RL-QMLAS suggests potential for broader application beyond current supervised learning tasks. Further exploration could involve extending this adaptability to noise-aware quantum devices and tackling increasingly complex quantum architectures. The scalable nature of the RL-based approach indicates promise for the frameworks to evolve alongside growing quantum computational capacities, potentially enabling richer and more efficient QML systems in the future.

In conclusion, this work marks a significant step towards automated QML via reinforcement learning, balancing circuit complexity against performance, and heralding a future where quantum computing can progressively integrate into varied scientific and practical domains.