Applications › Quantum and quantum-inspired machine learning
Applications · Machine learning
Quantum and quantum-inspired machine learning
Variational and kernel models on quantum hardware, quantum-inspired classical methods, and the question of when the quantum part is actually doing the work.
The problem
Quantum machine learning is the area of the field with the widest gap between claim and evidence. A model can be made quantum in several ways, and most of them do not help. The useful research question is narrow and specific: for which data, and which model class, does a quantum feature map or a variational circuit give something a good classical model does not already give?
We approach it from two directions. One is the conventional route — variational circuits, quantum kernels, and the optimisation problems inside training. The other is quantum-inspired: methods that borrow the structure without needing the hardware, which is where the fractional-memory work on neural architectures sits.
Variational quantum models and quantum kernel methods; combinatorial formulations of feature selection and model search; and quantum-inspired classical architectures — including Hamiltonian neural networks with fractional memory, aimed at long-range temporal dependence.
Very strong, and improving fast. Any claim in this area has to be measured against a properly tuned classical model on the same data, which is a discipline the field has not always observed. We try to observe it.
Published methods; benchmarking ongoing. The quantum-inspired side has produced results in Q1 venues. The hardware side is exploratory.
Time-series prediction with long memory; scientific machine learning where physical structure can be built into the model; feature selection on high-dimensional data.
A dataset and a task with a defensible classical baseline already in place. Without that baseline the comparison is not meaningful — and we would rather say so at the start than at the end.
Evidence
- “Fractional action Hamiltonian neural networks: long-term memory, sequential learning, and AI applications.” Neurocomputing 681, 133389 (2026). doi
- “Quantifying and quantizing non-Markovian effects in fiber links via fractional memory for predictive modeling and network adaptation.” Computer Networks 288, 112649 (2026). doi
Talk to us about this
We are candid about this area. If a classical model will serve you better, that is the advice you will get.
Last updated 14 September 2026.