Toward Joint Optimization of Circuit Depth and Training Data Size in Adaptively Grown Quantum Classifiers
Reimplemented Q-FLAIR finds no reliable link between MNIST training-set size and grown quantum circuit depth.
The authors reimplement Q-FLAIR's gate-by-gate circuit growth and exact stopping rule to test whether grown quantum classifiers scale predictably with training-set size under Caro et al.'s generalization bound. On full-resolution 784-pixel MNIST 3-versus-5 classification, with N from 2000 to 10000, both circuit size and test accuracy vary non-monotonically, and seed-to-seed variance is nearly as large as any trend. The empirical generalization gap stays within the bound in 14 of 15 runs but correlates only weakly with it (r = 0.12), so active gate count does not explain most observed variation.
- No predictable link between training-set size and grown circuit size
- MNIST 3-vs-5 tested at N from 2000 to 10000
- Circuit size and accuracy vary non-monotonically across seeds
- Caro et al. generalization bound holds in 14 of 15 runs
- Gap correlates only weakly with the bound (r = 0.12)
From the original article
1 min read at arxiv.orgBuilding a quantum model involves a tradeoff: how complex the circuit should be, and how much training data it needs. Caro et al. show that models with fewer trainable gates need less training data to generalize well.
By Saeefa Rubaiyet Nowmi · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.