Quantum Machine Learning Explained
The basic idea
Machine learning finds patterns in data, usually through heavy math on large arrays of numbers. Quantum machine learning asks whether qubits and quantum circuits could do some of that math faster or represent data in useful new ways.
Main approaches researchers explore
- Quantum circuits as models: a parameterized circuit is trained on a normal computer, similar to training a small neural network.
- Quantum kernels: quantum states are used to compare data points in a high dimensional space.
- Quantum speedups for linear algebra: some proposed methods promise faster math, but often assume data can be loaded into a quantum computer cheaply.
The hard parts
Loading classical data into qubits is difficult and can erase any speedup. Current machines are noisy, and training circuits can run into flat regions where learning stalls. Also, researchers have shown that some proposed quantum methods can be matched by new classical algorithms. This is a reminder to compare against the best normal approach.
What is realistic
Today's AI runs on classical hardware, and that is expected to remain true for most tasks. Quantum machine learning is most likely to help in niche cases, especially those involving quantum data from physics or chemistry. See use cases and common myths.
Crypto note
Be wary of tokens or tools that claim to use quantum AI to predict markets. The theme appears in memes like QUANTUM (QNT), which is a memecoin and not a quantum AI product.
The dequantization lesson
In 2018, as a young researcher, Ewin Tang showed that a celebrated quantum algorithm for recommendation systems could be matched in speed by a classical algorithm, under comparable assumptions about data access. Her result removed one of the strongest candidates for an exponential quantum machine learning speedup. The approach, now called dequantization, was extended to other quantum linear algebra methods. The lesson is not that quantum learning is dead, but that many promised speedups depend on assumptions that classical methods can also use.
The barren plateau problem
A 2018 Nature Communications paper by Google researchers showed that for a broad class of randomly initialized parameterized circuits, the chance of finding a useful training signal falls off exponentially as the number of qubits grows. Researchers call these flat regions barren plateaus. Many current proposals try to avoid them with careful circuit design, but it remains a key obstacle for variational methods.
Comparison
| Approach | Main promise | Main obstacle |
|---|---|---|
| Variational circuits | Run on small noisy machines | Barren plateaus, noise |
| Quantum kernels | New ways to compare data | Unproven benefit on real data |
| Quantum linear algebra | Possible large speedups | Data loading, dequantized in some cases |
| Quantum data learning | Learn from quantum experiments directly | Needs quantum sensors or simulators |
Common mistakes
- Comparing against a weak baseline. The right test is the best classical model with the same data.
- Using tiny datasets. Results on a handful of data points rarely scale.
- Mixing up quantum for AI with AI for quantum. Machine learning is already used to decode errors and tune quantum devices, which is a separate and more concrete use.
What is changing in 2026
Attention has shifted toward AI helping quantum, such as calibration and error decoding, as covered in real-time error correction. Claims of quantum advantage in learning remain rare and debated. As the HSBC and IBM bond result of September 2025 showed, back-tests with noisy hardware draw skepticism from experts. See bank pilots.
How to check this yourself
Ask: was the data loaded efficiently, what classical model was compared, and were the results repeated across datasets? Our hype checklist applies here too.
Sources and further reading
- Tang: a quantum-inspired classical algorithm for recommendation systems (arXiv)
- McClean et al. 2018: Barren plateaus in quantum neural network training landscapes (Nature Communications)
- PostQuantum.com: HSBC and IBM bond trading result
Checked 2026-10-09. Research and standards change often, so check the primary documents. Nothing here is financial advice. The QNT memecoin is independent of Quantinuum Ltd, the real company, and of every lab, company and standards body named on this page.
Frequently asked questions
Is quantum machine learning better than regular machine learning?
Not shown in general. Advantages are mostly theoretical or limited to special cases, and classical methods are very strong.
Does ChatGPT-style AI run on quantum computers?
No. Large language models run on classical hardware such as GPUs.
What is a quantum kernel?
It is a method that uses quantum states to measure similarity between data points, which a normal model can then use for classification.
Can quantum AI predict crypto prices?
No proven method exists. Treat any such claim as a warning sign.
What is a barren plateau?
It is a region of a training landscape where the learning signal is almost flat, so a model cannot tell which way to improve. It becomes more likely as circuits get larger.
What does dequantization mean?
It means finding a classical algorithm that matches a proposed quantum speedup, typically by using similar assumptions about how data can be sampled.
Can quantum computers help train large language models?
Not in any shown way. Models run on classical chips, and loading large datasets into qubits is a major obstacle.
Is AI helping quantum computers?
Yes, in areas like decoding errors and tuning hardware, which are practical and active research topics.
Keep reading
- Quantum Algorithms Explained for Beginners
What is a quantum algorithm? Learn how Shor's, Grover's and other quantum algorithms work in plain English, and which ones matter for cryptography and crypto. - Quantum Computing Use Cases: What Could It Actually Do?
From drug discovery to logistics and cryptography, here are the realistic use cases of quantum computing. - NISQ Explained: Noisy Intermediate-Scale Quantum Computers
What does NISQ mean? Learn why today's noisy, mid-size quantum computers are limited, what they can do, and how the field plans to move past them. - Quantum Computing Myths and Misconceptions
Common quantum computing myths, from 'it tries every answer at once' to 'Bitcoin breaks tomorrow'. Learn what is true, what is hype and what is still uncertain.
All Quantum computing guides | Back to top | Search the site
Main pages: Quantum computing explained | Quantum and crypto | Companies | Quantum news | Glossary