Research track · AI & machine learning
Machine-learning experimentation
Study quantum kernels and QNNs with honest baselines.
Why this horizon. Quantum kernels and neural networks face trainability limits (barren plateaus), and several proposed speed-ups were later matched by classical “dequantised” algorithms.
- The problem
- Claims about quantum machine learning are hard to evaluate without consistent baselines.
- Why quantum or hybrid methods may be relevant
- Quantum kernels and parameterized circuits are active research topics.
- Classical baseline
- Classical kernels and neural networks with matched data and tuning effort.
- What QAISS could provide
- Experiment templates, datasets, and comparison dashboards.
- What evidence would count as success
- Statistically sound comparisons on held-out data.
- Limitations & open questions
- Trainability and data-loading costs remain open research questions.
- Current status
- Planned No experiments have been published for this track yet.
Workflow on QAISS
From question to checkable result.
- STEP 1
Pick a dataset and a strong classical baseline (e.g. a tuned classical kernel method).
- STEP 2
Define the quantum feature map or circuit model.
- STEP 3
Estimate kernel entries or gradients from shots; watch the variance.
- STEP 4
Report accuracy and cost against the baseline, including negative results.
Sources
Read the literature.
- Supervised learning with quantum-enhanced feature spacesV. Havlíček, A. D. Córcoles, K. Temme et al. · Nature 567, 209–212 · 2019
- Barren plateaus in quantum neural network training landscapesJ. R. McClean, S. Boixo, V. N. Smelyanskiy et al. · Nature Communications 9, 4812 · 2018
- A quantum-inspired classical algorithm for recommendation systemsE. Tang · STOC 2019 · arXiv:1807.04271 · 2018
Cited for context. QAISS is not affiliated with the authors or publishers.
No quantum advantage, production savings, medical benefit, financial return, or security guarantee is claimed for this track.