Applications
Quantum Machine Learning (QML)
The intersection of quantum computing and machine learning, exploring whether quantum circuits can accelerate specific ML tasks.
Definition
Quantum machine learning is a research field exploring whether quantum computers can provide advantages for machine learning tasks, either by running quantum versions of classical ML algorithms or by designing entirely new quantum-native learning models.
In depth
Quantum machine learning combines quantum computing with machine learning, in several different ways. The data and the algorithm can each be classical or quantum, giving four settings. The most common, using quantum circuits to process classical data, includes quantum kernels, variational classifiers, and quantum neural networks. A second line uses classical learning to analyze quantum data, such as learning properties of a quantum state from measurements, and a third uses quantum algorithms to speed up linear algebra subroutines that machine learning relies on.
The evidence is more cautious than the headlines. Early proposals promised exponential speedups, such as for recommendation systems and principal component analysis, but Tang and others showed that classical algorithms with comparable data access can match them. Loading classical data into quantum states is expensive, and variational models suffer from barren plateaus. A 2021 study by Huang and coauthors showed that classical models with training data can match quantum ones on many tasks. A 2024 large-scale benchmark by Bowles, Ahmed, and Schuld found that simple classical models generally outperformed quantum ones on the tested data sets.
Where advantage seems most plausible is on quantum data, learning about quantum systems, where the input is already in a quantum form, and on tasks built around structure that classical computers cannot exploit, such as the discrete-logarithm-based classification of Liu, Arunachalam, and Temme in 2021. Research also explores quantum-inspired classical methods, such as tensor-network compression of neural networks. For practical machine learning on classical data, GPUs and large classical models remain dominant, and quantum machine learning is a research field with limited practical impact so far.
Technical Definition
Proven QML speedups generally apply only under specific, often idealized assumptions (such as efficient quantum data loading via QRAM), and as discussed in our Quantum vs AI comparison, no broad advantage has been demonstrated on real-world datasets using currently available hardware.
Visual Explanation: An Analogy
Think of quantum machine learning like an experimental new kind of paintbrush being tested by artists — promising in carefully controlled demonstrations, but not yet proven to produce better paintings than traditional tools when used in the messy, complex conditions of real artistic work.
Key facts
| Item | Detail |
|---|---|
| Four settings | Classical or quantum data combined with classical or quantum algorithms |
| Common models | Quantum kernels, variational classifiers, quantum neural networks, and quantum generative models |
| Dequantization | Tang (2018) showed classical algorithms matching quantum recommendation speedups |
| Benchmark (2024) | Bowles, Ahmed, and Schuld found classical models outperforming quantum ones on tested tasks |
| Open niche | Learning from quantum data and structured tasks with proven separations |
Where the idea comes from
Early quantum machine learning papers appeared in 2013 to 2014, including Lloyd, Mohseni, and Rebentrost's work. A wave of variational models followed in 2018 to 2021, with critical analyses and benchmarks from 2018 onward.
Real-World Use Cases
- An active academic research field spanning quantum neural networks, quantum kernel methods, and quantum versions of classical algorithms like k-means and SVM
- Distinct from (and currently less mature than) classical AI being used to help build better quantum hardware
How it differs from related ideas
- Quantum machine learning on classical data faces a data-loading bottleneck, while learning on quantum data does not.
- Quantum-inspired methods run on classical computers and borrow mathematical ideas.
- Classical deep learning scales with GPUs and large data, which quantum models do not yet match.
Common Misconceptions
- Quantum machine learning is not the same as 'quantum computers making AI smarter' in a broad sense — current proven advantages are narrow and problem-specific.
Frequently asked questions
What is quantum machine learning?
The use of quantum computers or quantum-inspired methods for learning tasks, including quantum kernels, quantum neural networks, and learning about quantum data.
Does it beat classical machine learning?
Not shown on practical tasks. Benchmarks generally find classical models performing as well or better.
Where might it help?
On quantum data, and on tasks where there is a proven separation based on structure that classical methods cannot exploit.
Related Terms
Keep exploring
- Algorithm: Quantum Kernel Estimation
- Algorithm: Quantum Support Vector Machine
- Algorithm: Quantum GAN (QGAN)
- Algorithm: Quantum Random Access Memory (QRAM)
- Hardware: IBM Heron (r1, r2, and r3)
- Industry: Quantum Computing, AI, and Data Centers
- Industry: Quantum Computing for Finance