AQER: a scalable and efficient data loader for digital quantum computers

Digital quantum computing promises to offer computational capabilities beyond the reach of classical systems, yet its capabilities are often challenged by scarce quantum resources. A critical bottleneck in this context is how to load classical or quantum data into quantum circuits efficiently. Approximate quantum loaders (AQLs) provide a viable solution to this problem by balancing fidelity and circuit complexity. However, most existing AQL methods are either heuristic or provide guarantees only for specific input types, and a general theoretical framework is still lacking. To address this gap, here we reformulate most AQL methods into a unified framework and establish information-theoretic bounds on their approximation error. Our analysis reveals that the achievable infidelity between the prepared state and target state scales linearly with the total entanglement entropy across subsystems when the loading circuit is applied to the target state. In light of this, we develop AQER, a scalable AQL method that constructs the loading circuit by systematically reducing entanglement in target states. We conduct systematic experiments to evaluate the effectiveness of AQER, using synthetic datasets, classical image and language datasets, and a quantum many-body state datasets with up to 50 qubits. The results show that AQER consistently outperforms existing methods in both accuracy and gate efficiency. Our work paves the way for scalable quantum data processing and real-world quantum computing applications.

💡 Research Summary

This paper tackles one of the most pressing bottlenecks in digital quantum computing: the efficient loading of classical or quantum data into a quantum processor. Exact state preparation is known to require resources that scale exponentially with the number of qubits, making it infeasible for near‑term devices. Approximate Quantum Loaders (AQLs) have emerged as a pragmatic alternative, trading a modest loss in fidelity for dramatically reduced circuit depth. However, existing AQL techniques fall into two broad families—tensor‑network (TN) based and circuit‑based—each of which either lacks rigorous performance guarantees or applies only to restricted classes of input states. Consequently, a unified theoretical understanding of the fundamental limits of AQLs has been missing.

The authors first introduce a unified optimization framework that captures virtually all known AQL methods. In this formulation the loader seeks a unitary U(θ; A) drawn from a fixed gate set such that the overlap |⟨v_target|U(θ; A)|ψ_product⟩|² is maximized, where |ψ_product⟩ is an easily preparable product state. Tensor‑network approaches correspond to incrementally extending U with local unitaries derived from an MPS or other low‑entanglement representation, while circuit‑based approaches either variationally optimize a fixed architecture or iteratively adjust both parameters and architecture.

With this common language the paper derives two information‑theoretic bounds on the achievable infidelity of any AQL (Theorem 3.1). The key quantity is an entanglement measure S(|ψ⟩)=∑_{i=1}^N S_i(|ψ⟩), the sum of single‑qubit Rényi‑2 entropies after the loading circuit is inverted. The lower bound f₁(S) and upper bound f₂(S) both scale linearly with S (or S/N when S is small), implying that the residual entanglement of the state U†|v_target⟩ directly controls the loading error. In the limit S→0, the bounds reduce to f₁(S)≈(ln 2/2)·N·S and f₂(S)≈(ln 2/2)·S, providing a clear, quantitative target for any loader: reduce the entanglement as much as possible.

Guided by this insight, the authors propose AQER (Approximate Quantum Entanglement Reduction). AQER proceeds in three stages:

1. Entanglement Reduction – Iteratively prepend two‑qubit gate blocks V_T(α) to the circuit, explicitly minimizing the entanglement measure of U†|v_target⟩. This step transforms a highly entangled target into a low‑entanglement intermediate state.

2. Product‑State Approximation – Approximate the resulting low‑entanglement state with a sequence of single‑qubit rotations R_Z(β_n) and R_Y(γ_n) applied to the all‑zero product state. Because the state is already weakly entangled, a shallow product‑state ansatz suffices.

3. Parameter Refinement – Fine‑tune all parameters θ=(α,β,γ) using a variational cost (infidelity) to obtain the final loader U_AQER(θ*). The entanglement‑reduction pre‑training supplies a good initialization, mitigating barren‑plateau problems and enabling rapid convergence even for 50‑qubit systems.

The experimental evaluation is extensive. The authors benchmark AQER on four families of data: (i) synthetic random quantum states, (ii) classical image datasets (MNIST, CIFAR) encoded as amplitude vectors, (iii) textual embeddings, and (iv) quantum many‑body states (e.g., 1‑D Ising and Heisenberg models) up to 50 qubits. Baselines include exact MPS loaders, variational circuit loaders, non‑variational local‑gate optimizers, and recent GAN‑based loaders. Across all tasks AQER consistently achieves lower infidelity (10–30 % improvement) and requires fewer two‑qubit gates (20–40 % reduction). The advantage is most pronounced for highly entangled many‑body states, where traditional methods often fail to reach acceptable fidelity. Moreover, AQER’s training converges within a few dozen epochs, and the authors report no observable barren‑plateau behavior.

All code and datasets are released on GitHub, ensuring reproducibility. The paper concludes by emphasizing that entanglement reduction is not merely a heuristic but a provably optimal strategy for approximate loading, and that AQER provides a practical, scalable implementation of this principle. Future directions include extending the entanglement metric to multipartitions, hybrid TN‑circuit architectures, and experimental validation on noisy intermediate‑scale quantum (NISQ) hardware with error mitigation.

In summary, this work delivers both a fundamental information‑theoretic characterization of AQL performance and a concrete algorithm that leverages this theory to achieve state‑of‑the‑art loading efficiency, paving the way for scalable quantum data processing in real‑world applications.