Artificial intelligence and quantum computing are two of the most talked-about technologies. Putting them together sounds unstoppable, but the details matter.
What quantum machine learning is
Quantum machine learning (QML) is the idea of using quantum computers to find patterns in data, train models or make predictions. Most proposals use a quantum circuit with adjustable settings, trained much like a neural network, with an ordinary computer doing the tuning.
The big obstacles
- ◆Loading data: modern AI learns from huge datasets. Getting large amounts of ordinary data into a quantum computer is slow, and can wipe out any speed-up.
- ◆Training difficulties: in 2018, researchers showed that many quantum models suffer from “barren plateaus”, flat landscapes where training can’t tell which way to improve.
- ◆Classical catch-up: also in 2018, Ewin Tang, then a student, showed that a celebrated quantum recommendation algorithm could be matched by a new classical one. Several other claimed speed-ups have since been matched in the same way.
- ◆Hardware limits: today’s noisy machines can run only small models, which ordinary computers handle easily.
Where it could help
QML looks most promising when the data is itself quantum, for example measurements from quantum sensors or results of chemistry simulations. Then there is no data-loading bottleneck, and quantum computers are naturally suited to the information.
The reverse is already happening
Ordinary AI is already helping quantum computing. Machine learning is used to calibrate qubits, design experiments and decode errors in error-correction schemes; in 2024, Google DeepMind described a neural network called AlphaQubit for exactly this job.
Practical advice
task_altKey takeaways
- check_circleData loading, training difficulties and classical catch-up limit QML today.
- check_circleQML is most promising for data that is already quantum.
- check_circleAI is already helping quantum computing, for example in error correction.
