Quantum Machine Learning: A Hands-on Tutorial for Machine Learning Practitioners and Researchers

TL;DR

This hands-on tutorial introduces quantum machine learning (QML) to AI practitioners by detailing foundational quantum computing concepts, core QML algorithms, and practical considerations such as trainability, generalization, and computational complexity, supplemented by code demos at https://qml-tutorial.github.io/. It surveys both near-term (NISQ) and fault-tolerant (FTQC) paradigms, covering quantum kernel methods, quantum neural networks (QNNs), and quantum transformer concepts, with emphasis on data encoding, kernel design, and linear algebra tools like block encoding and quantum singular value transformation (QSVT). The work provides concrete, executable pathways (via PennyLane and related libraries) for implementing QML techniques on real hardware or simulators, including MNIST classification and QK demonstrations, while highlighting read-in/read-out bottlenecks and strategies to mitigate them (e.g., shadow tomography, QRAM, data re-uploading). Overall, the tutorial articulates the current landscape of QML, clarifies theoretical foundations for expressivity and generalization, and sketches practical roadmaps for deploying QML in AI research and industry, acknowledging hardware constraints and the ongoing transition from FTQC to NISQ-enabled workflows.

Abstract

This tutorial intends to introduce readers with a background in AI to quantum machine learning (QML) -- a rapidly evolving field that seeks to leverage the power of quantum computers to reshape the landscape of machine learning. For self-consistency, this tutorial covers foundational principles, representative QML algorithms, their potential applications, and critical aspects such as trainability, generalization, and computational complexity. In addition, practical code demonstrations are provided in https://qml-tutorial.github.io/ to illustrate real-world implementations and facilitate hands-on learning. Together, these elements offer readers a comprehensive overview of the latest advancements in QML. By bridging the gap between classical machine learning and quantum computing, this tutorial serves as a valuable resource for those looking to engage with QML and explore the forefront of AI in the quantum era.

Figures (41)

• Figure 1: The paradigm between classical and quantum computing. The mechanisms between classical and quantum computing are very similar, where both of them involve input, computation, and output. In classical computing, the input refers to a bit-string, the computation part refers to the digital logic circuits, and the output also refers to a bit-string. In quantum computing, the input is a single- or multi-qubit state. The computation involves quantum circuits. And the output of quantum computers requires quantum measurement, which aims to extract information from the quantum world to the classical world.
• Figure 2: Different research directions in QML. QML can be categorized into four types based on the interplay of quantum (Q) and classical (C) systems: Q for Q (quantum algorithms for quantum data), Q for C (quantum algorithms for classical data), C for Q (classical algorithms for quantum data), and C for C (classical algorithms for classical data). This tutorial primarily focuses on the Q for C category. Beyond the role of the learner and system, QML can also be classified by learning tasks (discriminative and generative learning), learning paradigms (supervised, unsupervised, and reinforcement learning), and diverse applications such as chemistry, computer vision, power systems, logistics, finance, and healthcare.
• Figure 3: Common quantum architectures and roadmaps from different quantum companies.
• Figure 4: Mechanisms of DNNs and QNNs. Both DNNs and QNNs follow an iterative approach. At each iteration, they take input data, process it through multiple layers, and produce an output prediction. The key difference between DNNs and QNNs is the way of implementing their learning models.
• Figure 5: The learnability of quantum machine learning models.

Theorems & Definitions (91)

• Example 2.1
• Definition 2.2
• Example 2.3
• Definition 2.4
• Definition 2.5
• Example 2.6
• Example 2.7
• Definition 2.9: CPTP map
• Definition 2.10: Depolarization channel
• Example 2.11