ECAI25 Tutorial - A Gentle Introduction to Quantum Machine Learning

Quantum Machine Learning is a subject that has to navigate the huge expectations of its parent disciplines. Today, there are many applications of quantum machine learning in the literature, with various claims over their performance vs. more classical machine learning approaches. Yet, with quantum technology advancing at a tremendous speed, there is a need for AI researchers to understand what this new family of machine learning methods can, and cannot do. This tutorial seeks to illustrate, with a high emphasis on practical examples using the Qiskit framework, what quantum machine learning has to offer, and give an introduction to the general process of how to train simple quantum machine learning models.

The tutorial will comprise the following subject matter:

• Gentle introduction of quantum computing and its formal foundations
• Representing, and handling data on a Quantum Computer
• Techniques for preprocessing, and transforming tabular data QML methods
• Supervised training of QML models, and performing inference
• Characterizing Performance in ML vs. QML
• Discuss current challenges and research directions to give an outlook on where participants could contribute to this emerging area of research

Specifically, we will focus on the family of supervised QML methods that leverage variational techniques. In the literature, this family of methods is often referred to as Quantum Neural Networks (QNNs).

After an introduction to the general domain of quantum computing, the tutorial will guide its participants through annotated Jupyter notebooks that capture:

• Implementing a "dumb" rule-based model that acts as a classifier (of sorts) on the Titanic dataset. This will illustrate how to encode the features Gender and Age for a quantum computer, and implement a simple quantum program to predict ``Survived'' (true) if the passenger is female and/or under 18, and false otherwise. This will highlight basic fundamental concepts of constructing and executing a quantum program with a standard introductory-level machine learning dataset.
• We will then progress to training a QNN on the same dataset and discuss different data preparation strategies, and quantum architecture decisions that influence model training and performance.
• Finally, we will discuss how to incorporate noise (i.e. NISQ errors), and (time and presentation environment permitting) how to deploy our model to a cloud-connected quantum device for inference on a real quantum computer. This might need to be a pre-baked demo by the presenters as it can be hard to predict device availability (specifically queue times) over the Cloud.

This structure would allow participants to then progress at their own pace onto more advanced topics and tutorials available on the web. As such, our objective is to get them started, and have a little confidence in managing their own learning into more advanced concepts and scenarios.