arXiv:2610.17836
A Tutorial on Quantum Amplitude Embedding with Data Re-Uploading
Hyojun Ahn, Joongheon Kim
Abstract
Amplitude embedding lets a variational quantum model read many classical features with few qubits, but the model output is then only a quadratic form in the normalized input. Data re-uploading, which encodes the input repeatedly between trainable blocks, can be used to enrich such a model. Because every additional upload of an amplitude embedding repeats a costly state preparation, the gain that an upload buys has to be known before the gates are spent. This tutorial thus addresses the problem of explaining what data re-uploading adds to an amplitude-embedded model and how large the gain is. We express sequential re-uploading and parallel re-uploading with two copies in one notation and show that each upload raises the reachable polynomial order by two. Under one trainable circuit family and one budget of trainable parameters, simulation results show that re-uploading raises the test coefficient of determination from 0.689 to between 0.883 and 0.975 on targets of order four and from 0.337 to between 0.642 and 0.783 on targets of order six, on all ten seeds. On a quadratic target and on handwritten digits, one upload remains the best choice, and every additional upload lowers the gate error that the classifier tolerates. These results help designers match the number of uploads to the polynomial order of the task at hand.
AI Summary
The paper asks when repeated amplitude embedding improves a variational quantum model enough to justify additional state preparations. It reviews the single-upload restriction $f(\mathbf{x})=\tilde{\mathbf{x}}^T M\tilde{\mathbf{x}}$ and compares sequential re-uploading on one register with parallel encoding into multiple registers. The analysis treats sequential uploads as angle encodings in the state-preparation angles: $L$ uploads extend the accessible frequencies to $L$. It connects this to degree-$2L$ targets, arguing that such targets generally require at least $L$ uploads. Parallel encoding into $k$ copies gives outputs homogeneous of degree $2k$ in the normalized input. For the specified preparation of nonnegative inputs, each upload costs $2^n-2$ CNOTs. Simulations hold the trainable ansatz family, parameter budget, and training procedure fixed across schemes. On synthetic degree-four targets, test $R^2$ increases from 0.689 with one upload to 0.883–0.918 with two to four sequential uploads and 0.975 with two copies. On degree-six targets, the corresponding values are 0.337, 0.642–0.783, and 0.771; every repeated-encoding model beats one upload on all ten seeds. One upload performs best on quadratic targets and handwritten digits: digit accuracy with 1,297 training images falls from 95.9% to 88.5% with four uploads. A global-depolarizing, finite-shot evaluation also finds reduced gate-error tolerance with additional uploads. The stated scope is simulations of three to twelve qubits, nonnegative inputs, one ansatz family, and noiseless training. The proposed design rule is to match upload count to target polynomial order.
Central claim may not hold: Sequential amplitude re-uploading can produce nonpolynomial functions of normalized features, so the preparation-angle frequency bound does not establish the abstract’s general claim that each upload raises reachable polynomial order by two. The recommendation to match sequential upload count to target polynomial degree lacks a reachability argument: a necessary frequency condition and improvements on selected numerical targets do not establish sufficiency.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 6 pages, 6 figures, 3 tables. Submitted to ICEIC 2027