I did computing 7 years ago, right after my master's degree. At that time, the field was full of excitement but also doubt. Today, quantum computing stands out as an emerging technology, alongside HPCs and AI.
Attention has shifted from hardware-only research and discussion to applications, software, and algorithms. Quantum is a tool that can be used across a variety of fields rather than a single field. Another promising, yet not fully understood application of quantum computers is quantum machine learning.
Quantum machine learning (QML) has become a catch-all term in the last few years. One of the earliest and most important appearances of QML was in 2013, when Google and NASA established the Quantum Artificial Intelligence Lab, which was tasked with exploring how quantum computers can be used in machine learning applications. Since then, the term QML has appeared in research papers, startups, and conference discussions, often with very different meanings.
In some cases, it refers to using quantum computers to accelerate machine learning. In others, it describes classical algorithms inspired by quantum physics. And sometimes, it just means running standard ML workflows on non-standard hardware.
So even I, someone who works and researches quantum computers, was very confused at first… I bet that the first question of many people when they hear “Quantum Machine Learning” is what exactly, makes a quantum machine learn. quantum?
Answering this question is why I decided to write this article! The short answer is not speed, and it's not neural networks, and it's not vague indications of “quantum gain.” At its core, quantum machine learning is defined as how information is represented, transformed, and read. In QML, that is done using the laws of quantum mechanics instead of classical calculus.
This article aims to clarify those differences, separate substance from hype, and provide a clean conceptual foundation for the rest of the series. I plan to write about testing the theory of QML, as well as some of the near-term research results and applications.
Machine Learning Before “Quantum”
Before we get all quantum, let's go back. Stripped of its modern elements, machine learning is about learning a map from input to output using data. Whether the model is a linear regressor, a kernel method, or a deep neural network, the structure is very similar:
- Data is represented by numbers (vectors, matrices, tensors).
- A parameterized model replaces that data.
- The parameters are adjusted by optimizing the cost function.
- The model is tested statistically on new samples.
Neural networks, GPUs, and large datasets are implementation options and do not define features. This deduction is important because it allows us to ask a specific question:
What changes when the the data and the model reside in quantum space?
Quantum Mechanics Intrudes
Quantum machine learning becomes quantum when quantum information is the integrated substrate. This can be seen in three ways.
1. Data is represented as quantum states.
In classical machine learning models, data is represented as bit numbers or floating points. In contrast, quantum machine learning uses quantum states, which are complex vectors that obey the laws of quantum mechanics. These regions are often described by density matrices, and their transformations are represented by unity matrices.
As a result, we write data with amplitudes of a more complex value than is possible, and regions can exist with height.
This does it not it means that all old data is already highly compressed or accessible. Loading data into quantum states is often expensive, and extracting information from them is fundamentally limited by measurement.
So, the important point is that the model works on quantum states, not classical numbers.
2. Models of Quantum Evolutions
Classic ML models apply functions to the data. Quantum ML models use quantum operations (usually integral transformations) on quantum channels. In practice, most QML models are built from parameterized quantum circuits. These states are a sequence of quantum gates, which are the basic functions that switch quantum states. The parameters of these quantum gates are tuned during training, similar to adjusting the weights in a neural network in classical machine learning.
Basically, what happens in these examples is that we start with the state of the system, represented in a matrix (we'll call it Hamiltonian, to be precise), and then the gates we put into the system will tell us how that system changes (changes) over a period of time. That variable determines the behavior of the model.
As a result, quantum models explore a hypothesis space that is structurally different than classical models, even if the training loop appears superficially identical.
3. Evaluation Is Part of the Learning Process
In classic ML, reading the output of a model is trivial and doesn't affect the state or behavior of the model at all (unless we intentionally make it so). In quantum ML, however, scaling is a possibility and it harms the state. This has a huge impact on the system. The output is determined by the frequency of the circuit, called 'shots.' Here, 'shooting' means running the quantum circuit multiple times to measure the result, since quantum measurement is a possibility.
The gradients (how the parameters of the guidelines are updated during training) are estimated mathematically from these measurements rather than being precisely computed as in classical machine learning. As a result, training costs are often dominated by sampling noise from these repeated measurements, rather than by calculation alone.
In other words, uncertainty is built into the model itself. Any serious discussion of QML must address the fact that learning is possible by using measure, not after you.
What does No Make QML Quantum
Quantum computing and QML, in particular, generate hype and misunderstanding. Many things called “quantum machine learning” today are quantum in name only, for example:
- Classical ML algorithms run on quantum hardware without the rational use of quantum states.
- “Quantum-inspired” methods that are absolutely state-of-the-art.
- Pipelines are hybrids where a quantum component can be removed without changing the behavior or performance of the model.
If you ever meet someone talking about QML and you're not sure what model they're talking about, a good rule of thumb to follow is to ask:
“Can I replace the quantum part with a classical one without changing the mathematical structure of the model?”
If yes or maybe, the method is probably not fundamentally quantum. This work may be important, but it falls outside the core of quantum machine learning.
Where is QML Today?
When discussing quantum computing, remember that current hardware is noisy, small, and resourceful. Because of this:
- There is no standard, proven quantum advantage for machine learning jobs today.
- Many QML models resemble kernel methods over deep networks.
- Data and audio loading often dominate performance.
This is not a failure of the field; this is where quantum computing stands right now. Most QML research is now exploring: mapping model classes, understanding quantum learning theory, and identifying where quantum architecture can be important.
Why Quantum Machine Learning Is Still Worth Studying
If near-term speedups are not possible, why pursue QML at all?
QML forces us to rethink fundamental questions about machine learning and quantum computing. We need to answer what it means to learn from quantum data, how noise affects optimization, and what classes of models exist in quantum systems but not in classical systems.
Quantum machine learning is less about the efficiency of classical ML today and more about expanding the space of what “learning” can mean in the quantum world.
This is important because scientific and technological progress starts in new ways. Even if the hardware isn't ready yet, testing QML prepares us for better hardware in the future.
Final thoughts and what's next
Advances in quantum computing are accelerating. Hardware companies are racing to build a fault-tolerant quantum computer. A quantum computer that uses the full power of quantum mechanics. Software and application companies are exploring problems that quantum computing can logically solve.
That said, today's quantum computers cannot run a life-sized operating system, let alone a complex machine learning model. Nevertheless, the promise of quantum computing efficiency in machine learning is quite interesting and should be explored now, along with hardware developments.
In this article, I focus on the definitions and boundaries of quantum machine learning to pave the way for future articles that will explore:
- How primitive data is embedded in quantum states.
- Various quantum models and their limitations.
- Quantum kernels and feature spaces.
- Optimization challenges in noisy quantum systems.
- Where the quantum advantage may appear clearly.
Before asking whether quantum machine learning is useful, we need to clarify what it actually is. The further away we are from the conversation, the less likely we are to move forward.
