Quantum AI Pattern Catalogue

The Quantum AI Pattern Catalogue provides a set of architectural solutions for quantum AI systems. Specifically, these system-level solutions aim to enhance system quality by addressing the inherent constraints of contemporary quantum hardware.

Motivation

Exploiting quantum computing technology for artificial intelligence (AI) systems has recently attracted increasing attention as a potential way to enhance their performance and possibly realise quantum advantage. The performance enhancement can manifest as faster training, reduced inference time, adversarial or noise robustness, or reduced memory consumption. However, moving beyond proof-of-concept or simulations to develop practical applications of these systems, while ensuring high software quality, faces significant challenges due to the limitations of quantum hardware and the underdeveloped knowledge base in software engineering for such systems.

Numerous studies have explored the physical, mathematical, and algorithmic foundations of quantum machine learning (QML). However, only a limited number of works have focused on the practical design of applications for existing quantum computers, particularly for Noisy Intermediate-Scale Quantum (NISQ) devices. NISQ computers, as defined by Preskill (2018), are quantum computing devices that are characterized by a limited number of qubits, limited depth of quantum cirquits and significant susceptibility to noise. The inherent limitations of NISQ hardware, coupled with non-functional requirements, such as performance, reliability, security, and usability, defined by software quality attributes, pose significant challenges in translating quantum algorithms into deployable and high-quality real-world applications.

The challenges outlined above are a central focus of quantum software engineering. The creation of high-quality quantum software and development of quantum software engineering are crucial steps in the overall progress of quantum computing, as asserted in “The Talavera Manifesto” (Piattini et al. 2020). Solutions developed during the quantum software engineering activities often take the form of formalised architectural and design patterns. The architectural patterns represent reusable practices and solutions for addressing recurring problems in software architecture (Q. Lu et al. 2024).

Overview of Quantum AI Pattern Catalogue

Reference architecture

We adopt a pattern-oriented approach and build up a Quantum AI Pattern Catalogue for operationalizing quantum AI from a system perspective. Architectural patterns provide foundational building blocks that a reference architecture integrates to address specific system concerns.

Figure 1. A multi-layered reference architecture for a quantum-enhanced AI system.

The reference architecture of quantum AI systems, shown in Figure 1, is similar to the classical one described in (Q. Lu et al. 2024) and is divided into several key layers, each playing a crucial role in the overall design of the quantum AI system. We present these layers as follows:

System layer. This layer of the quantum AI system architecture is responsible for executing the trained models and making predictions or decisions based on input data. It comprises non-AI components, such as the user interface, business logic, and data preprocessing tools, as well as AI components, including both classical and quantum inference engines. The quantum inference engine can be implemented using various design approaches, each featuring different quantum circuit architectures, a brief overview of which is provided here.

Operation layer. This layer in the architecture is responsible for the ongoing management, maintenance, and monitoring. This layer supports the seamless operation and maintenance of AI systems through the use of DevOps practices and monitoring tools.

Supply chain layer. This layer encompasses various components essential for managing and optimizing supply chain operations using both classical and quantum computing techniques. These components collaborate to collect, process, and analyze data, ultimately enhancing decision-making and efficiency in supply chain management.

Hardware layer. This layer includes computational resources necessary for implementing quantum AI systems, such as quantum computers, client hardware and server hardware.

Architectural patterns

Each pattern presents a high-level reusable solution to a problem commonly occurring within a given context in software development. Patterns are formally represented using a template that includes the following components: name (with an optional graphical representation), summary, description of the associated problem, solution provided by the pattern, benefits, drawbacks, and known uses. We classify the identified architectural patterns into two categories – quantum-classic split and quantum middleware architectural patterns.

Quantum-classic split architectural patterns

All quantum-classical split patterns are defined within the context of hybrid classical-quantum AI systems. For practical AI software applications, a hybrid quantum-classical software architecture is essential and, arguably, necessary. The patterns focus on integrating quantum components for data processing and accelerating computations with dedicated quantum hardware. In the course of the systemtic mapping study, we have identified the following patterns:

  • Quantum monolith: A single, unified quantum component.
  • Multi-layer: Integrates quantum processing as multiple layers of the architecture.
  • Quantum feature engineering: Quantum components that handle the initial stages of data processing and implement methods for extracting features from data
  • Quantum head: A quantum component that handles the last stages of the inference process.
  • Quanvolution: A quantum analog of convolutional layers in neural networks.
  • Intermediate quantum Layer: A quantum layer situated between classical layers.
  • Quantum accelerator: Hardware specifically designed to accelerate numerical computations.

Figure 2. Simple graphical representations of the quantum-classic split architectural patterns.

Quantum middleware architectural patterns

Quantum middleware architectural patterns are defined within the context of the interaction of quantum components and classical software. The latter includes operating systems and network infrastructure. Many of these interactions imply the existence of a middleware software layer that can organise the communication, orchestration, and scheduling of quantum components and workflows. The corresponding identified patterns are:

Our paper

Mykhailo Klymenko, Thong Hoang, Xiwei Xu, Zhenchang Xing, Muhammad Usman, Qinghua Lu, Liming Zhu, “Architectural patterns for designing quantum artificial intelligence systems,” Journal of Systems and Software, 227, 112456 (2025)

Related projects