Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2503.17113 · EPJ Quantum Technology

Fast quantum amplitude encoding of typical classical data

Vittorio Pagni, Sigurd Huber, Michael Epping, Michael Felderer

Published Feb 5, 2026• 6 citations
arXivHTMLPDF

Abstract

Abstract We present an improved version of a quantum amplitude encoding scheme that encodes the N entries of a unit classical vector $\boldsymbol{v}=(v_{1},..,v_{N})$ v = ( v 1 , . . , v N ) into the amplitudes of a quantum state. Our approach has a quadratic speed-up with respect to the original one. We also describe several generalizations, including to complex entries of the input vector and a parameter M that determines the parallelization. The number of qubits required for the state preparation scales as $\mathcal{O}(M\log N)$ O ( M log N ) . The runtime, which depends on the data density ρ and on the parallelization paramater M , scales as $\mathcal{O}(\frac{1}{\sqrt{\rho}}\frac{N}{M}\log (M+1))$ O ( 1 ρ N M log ( M + 1 ) ) , which in the most parallel version ( $M=N$ M = N ) is always less or equal than $\mathcal{O}(\sqrt{N}\log N)$ O ( N log N ) . By analysing the data density, we prove that the average runtime is $\mathcal{O}(\log ^{1.5} N)$ O ( log 1.5 N ) for input vectors that are uniformly sampled on the N -sphere. We present numerical evidence that this favourable runtime behaviour also holds for real-world data, such as radar satellite images. This is promising as it allows for an input-to-output advantage of the quantum Fourier transform.

DOI
10.1140/epjqt/s40507-026-00473-3

Related papers

Approximate Amplitude Encoding with the Adaptive Interpolating Quantum Transform

Gekko Budiutama, Shunsuke Daimon, Xinchi Huang +2 more

Amplitude encoding of real-world data on quantum computers is often the workflow bottleneck: direct amplitude encoding scales poorly with input size and can offset any speedups in subsequent processing. Fourier-based sparse amplitude encoding lowers cost by retaining only a small subset of dominant coefficients, but its fixed, non-adaptive basis leads to significant information loss. In this work, we replace the Fourier transform with the adaptive interpolating quantum transform (AIQT) in the sparse amplitude encoding workflow. The AIQT learns a data-adapted basis that concentrates information into a small number of coefficients. Consequently, at matched sparsity, the AIQT retains more information and achieves lower reconstruction error compared to the Fourier baseline. On financial time-series data, the AIQT reduces reconstruction error by 40% relative to the Fourier baseline, and on image datasets the reduction is up to 50% at the same sparsity level, with nearly identical encoding gate cost. Crucially, the approach preserves the efficiency of Fourier-based methods: the AIQT is built on the structure of the quantum Fourier transform circuit. Its gate count scales quadratically with the number of qubits, while classical evaluation can be carried out in quasilinear time. In addition, the AIQT is trained without labels and does not require sampling from quantum hardware or a simulator, removing a major bottleneck in data-driven amplitude-encoding methods.

quant-pharXiv:2603.03803•Mar 5, 2026•0 citations

Hybrid Quantum State Preparation via Data Compression

Emad Rezaei Fard Boosari, Maryam Afsary

Quantum state preparation (QSP) for a general $n$-qubit state requires $O(2^n)$ CNOT gates and circuit depth, making exact amplitude encoding (EAE) impractical for near-term quantum hardware. We introduce an ancilla-free hybrid classical-quantum strategy that reduces this cost to $O(poly(n))$ for a broad class of compressible data. The method first applies a classical compression step to obtain a $d$-sparse representation of the input, loads this sparse vector using a sparse-state preparation routine, and then reconstructs the target state through a polynomial-depth quantum inverse transform. We evaluate the framework on synthetic benchmark signals and real biomedical time series using Fourier and Haar transforms, demonstrating substantial reductions in CNOT counts and circuit depth compared to EAE, together with competitive performance relative to the Fourier Series Loader (FSL). The quantum simulation results show that combining classical data compression with quantum decompression provides a scalable framework for efficient QSP, reducing quantum overhead without requiring variational training or ancillary registers.

quant-pharXiv:2512.01798•Dec 2, 2025•4 citations

Sublinear Classical-to-Quantum Data Encoding Using nnn-Toffoli Gates

Vittorio Pagni, Gary Schmiedinghoff, Kevin Lively +2 more

Quantum state preparation, also known as encoding or embedding, is a crucial initial step in many quantum algorithms and often constrains theoretical quantum speedup in fields such as quantum machine learning and linear equation solvers. One common strategy is amplitude encoding, which embeds a classical input vector of size$\mathrm{N}=2^{\mathrm{n}}$in the amplitudes of an n-qubit register. For arbitrary vectors, the circuit depth typically scales linearly with the input size N, rapidly becoming unfeasible on near-term hardware. We propose a general-purpose procedure with sublinear average depth in N, increasing the window of utility. Our amplitude encoding method encodes arbitrary complex vectors of size$\mathrm{N}=2^{n}$at any desired binary precision using a register with$n$qubits plus 2 ancillas and a sublinear number of multi-controlled NOT (MCX) gates, at the cost of a probabilistic success rate proportional to the sparsity of the encoded data. The core idea of our procedure is to construct an isomorphism between target states and hypercube graphs, in which specific reflections correspond to MCX gates. This reformulates the state preparation problem in terms of permutations and binary addition. The use of MCX gates as fundamental operations makes this approach particularly suitable for quantum platforms such as ion traps and neutral atom devices. This geometrical perspective paves the way for more gate-efficient algorithms suitable for nearterm hardware applications.

arXiv:2505.06054•IEEE International Conference on Quantum Computing and Engineering (QCE)•Aug 30, 2025•3 citations

Q-MERGE: Parallelising Quantum State Preparation for Large-Scale Classical Data

Archie Butterworth, Jens Renders, Jingbo Wang

Quantum processing of classical data fundamentally relies on efficiently mapping classical datasets onto the amplitudes of quantum states. Preparing large amplitude-encoded states, however, remains a major bottleneck in quantum computing. In this paper, we introduce a novel state-preparation framework we call Q-MERGE, that addresses this bottleneck by partitioning a large target state into $M$ $n$-qubit segments, preparing these segments independently and in parallel, and then coherently combining them into a single amplitude-encoded state using SELECT-SWAP operations and measurement. Q-MERGE is agnostic to the segment-level preparation method, allowing existing techniques to be applied to smaller subproblems while providing a tunable trade-off between circuit depth and qubit count. With mid-circuit measurement and preparation-register reuse, the required ancilla qubits can be reduced from O(Mn) to O(n). Applied to a real-world $32 x 1024$ ultrasound dataset, Q-MERGE achieves an infidelity of $3.5 x 10^{-9}$, compared with $6.2 x 10^{-3}$ for direct preparation using the same underlying method, a more than six orders of magnitude improvement at equal CNOT circuit depth. We demonstrate its feasibility experimentally with a 53 qubit circuit comprised of 954 1-qubit gates and 478 2-qubit gates, on the Quantinuum System Model H2 trapped-ion quantum computer. The prepared state is validated using shadow-overlap tomography. Numerical analysis of Haar-random states indicates consistent success probability for up to $M=10^7$ segments, supporting the scalability of Q-MERGE for encoding massive classical datasets.

quant-pharXiv:2610.04247•Oct 6, 2026•0 citations

Limitations of Amplitude Encoding on Quantum Classification

Xin Wang, Yabo Wang, Qi, Bo +1 more

It remains unclear whether quantum machine learning (QML) has real advantages when dealing with practical and meaningful tasks. Encoding classical data into quantum states is one of the key steps in QML. Amplitude encoding has been widely used owing to its remarkable efficiency in encoding a number of $2^{n}$ classical data into $n$ qubits simultaneously. However, the theoretical impact of amplitude encoding on QML has not been thoroughly investigated. In this work we prove that under some broad and typical data assumptions, the average of encoded quantum states via amplitude encoding tends to concentrate towards a specific state. This concentration phenomenon severely constrains the capability of quantum classifiers as it leads to a loss barrier phenomenon, namely, the loss function has a lower bound that cannot be improved by any optimization algorithm. In addition, via numerical simulations, we reveal a counterintuitive phenomenon of amplitude encoding: as the amount of training data increases, the training error may increase rather than decrease, leading to reduced decrease in prediction accuracy on new data. Our results highlight the limitations of amplitude encoding in QML and indicate that more efforts should be devoted to finding more efficient encoding strategies to unlock the full potential of QML.

arXiv:2503.01545•Mar 3, 2025•13 citations

A Generative model to represent real time classical data as quantum state

Indraneel Borgohain, Charles D. Conner, Bharat S. Rawal

Existence of quantum systems in superposition states gives an exponential advantage over classical system in both time and space. In this paper we will look at reconstructing real time classical data as a pure quantum state. Each classical data sample eg. each pixel location in an images is indexed by a$\mathbf{n}$bit string representing one of the basis superposition states. The amplitude of the state which represents probability of measuring the state can be encoded as the normalized value of the sample at that location. A parameterized circuit[2] is used to prepare the quantum state as superposition of basis states from initial ground state. Each basis state corresponds to a sample classical data (eg. a pixel in case of an image). Quantum amplitude estimation based on Grovers algorithm[15] is used to estimate the amplitudes of each of these states. The parameters of the circuit is adjusted to minimize a loss function using classical gradient descent algorithm. The normalized amplitude at the output of our circuit is multiplied by a normalization factor that can also be tuned to represent the original data sample. This enables us to use the original real time data in our loss function computation. Section 1 of the paper gives a brief introduction. Section II covers the related works. Section III talks about amplitude encoding. Section IV describes quantum state preparation using supervised Learning. Section V describes implementation of the algorithms and finally Section VI gives the final conclusion of the research paper along with a result from the algorithm.

International Conference on Microwave, Antenna and Communication (MAC)•Mar 24, 2023•0 citations
Abstract We present an improved version of a quantum amplitude encoding scheme that encodes the N entries of a unit classical vector v=(v1,..,vN)\boldsymbol{v}=(v_{1},..,v_{N})v=(v1​,..,vN​) v = ( v 1 , . . , v N ) into the amplitudes of a quantum state. Our approach has a quadratic speed-up with respect to the original one. We also describe several generalizations, including to complex entries of the input vector and a parameter M that determines the parallelization. The number of qubits required for the state preparation scales as O(Mlog⁡N)\mathcal{O}(M\log N)O(MlogN) O ( M log N ) . The runtime, which depends on the data density ρ and on the parallelization paramater M , scales as O(1ρNMlog⁡(M+1))\mathcal{O}(\frac{1}{\sqrt{\rho}}\frac{N}{M}\log (M+1))O(ρ​1​MN​log(M+1)) O ( 1 ρ N M log ( M + 1 ) ) , which in the most parallel version ( M=NM=NM=N M = N ) is always less or equal than O(Nlog⁡N)\mathcal{O}(\sqrt{N}\log N)O(N​logN) O ( N log N ) . By analysing the data density, we prove that the average runtime is O(log⁡1.5N)\mathcal{O}(\log ^{1.5} N)O(log1.5N) O ( log 1.5 N ) for input vectors that are uniformly sampled on the N -sphere. We present numerical evidence that this favourable runtime behaviour also holds for real-world data, such as radar satellite images. This is promising as it allows for an input-to-output advantage of the quantum Fourier transform.
Amplitude encoding of real-world data on quantum computers is often the workflow bottleneck: direct amplitude encoding scales poorly with input size and can offset any speedups in subsequent processing. Fourier-based sparse amplitude encoding lowers cost by retaining only a small subset of dominant coefficients, but its fixed, non-adaptive basis leads to significant information loss. In this work, we replace the Fourier transform with the adaptive interpolating quantum transform (AIQT) in the sparse amplitude encoding workflow. The AIQT learns a data-adapted basis that concentrates information into a small number of coefficients. Consequently, at matched sparsity, the AIQT retains more information and achieves lower reconstruction error compared to the Fourier baseline. On financial time-series data, the AIQT reduces reconstruction error by 40% relative to the Fourier baseline, and on image datasets the reduction is up to 50% at the same sparsity level, with nearly identical encoding gate cost. Crucially, the approach preserves the efficiency of Fourier-based methods: the AIQT is built on the structure of the quantum Fourier transform circuit. Its gate count scales quadratically with the number of qubits, while classical evaluation can be carried out in quasilinear time. In addition, the AIQT is trained without labels and does not require sampling from quantum hardware or a simulator, removing a major bottleneck in data-driven amplitude-encoding methods.
Quantum state preparation (QSP) for a general nnn-qubit state requires O(2n)O(2^n)O(2n) CNOT gates and circuit depth, making exact amplitude encoding (EAE) impractical for near-term quantum hardware. We introduce an ancilla-free hybrid classical-quantum strategy that reduces this cost to O(poly(n))O(poly(n))O(poly(n)) for a broad class of compressible data. The method first applies a classical compression step to obtain a ddd-sparse representation of the input, loads this sparse vector using a sparse-state preparation routine, and then reconstructs the target state through a polynomial-depth quantum inverse transform. We evaluate the framework on synthetic benchmark signals and real biomedical time series using Fourier and Haar transforms, demonstrating substantial reductions in CNOT counts and circuit depth compared to EAE, together with competitive performance relative to the Fourier Series Loader (FSL). The quantum simulation results show that combining classical data compression with quantum decompression provides a scalable framework for efficient QSP, reducing quantum overhead without requiring variational training or ancillary registers.
Quantum state preparation, also known as encoding or embedding, is a crucial initial step in many quantum algorithms and often constrains theoretical quantum speedup in fields such as quantum machine learning and linear equation solvers. One common strategy is amplitude encoding, which embeds a classical input vector of sizeN=2n\mathrm{N}=2^{\mathrm{n}}N=2nin the amplitudes of an n-qubit register. For arbitrary vectors, the circuit depth typically scales linearly with the input size N, rapidly becoming unfeasible on near-term hardware. We propose a general-purpose procedure with sublinear average depth in N, increasing the window of utility. Our amplitude encoding method encodes arbitrary complex vectors of sizeN=2n\mathrm{N}=2^{n}N=2nat any desired binary precision using a register withnnnqubits plus 2 ancillas and a sublinear number of multi-controlled NOT (MCX) gates, at the cost of a probabilistic success rate proportional to the sparsity of the encoded data. The core idea of our procedure is to construct an isomorphism between target states and hypercube graphs, in which specific reflections correspond to MCX gates. This reformulates the state preparation problem in terms of permutations and binary addition. The use of MCX gates as fundamental operations makes this approach particularly suitable for quantum platforms such as ion traps and neutral atom devices. This geometrical perspective paves the way for more gate-efficient algorithms suitable for nearterm hardware applications.
Quantum processing of classical data fundamentally relies on efficiently mapping classical datasets onto the amplitudes of quantum states. Preparing large amplitude-encoded states, however, remains a major bottleneck in quantum computing. In this paper, we introduce a novel state-preparation framework we call Q-MERGE, that addresses this bottleneck by partitioning a large target state into MMM nnn-qubit segments, preparing these segments independently and in parallel, and then coherently combining them into a single amplitude-encoded state using SELECT-SWAP operations and measurement. Q-MERGE is agnostic to the segment-level preparation method, allowing existing techniques to be applied to smaller subproblems while providing a tunable trade-off between circuit depth and qubit count. With mid-circuit measurement and preparation-register reuse, the required ancilla qubits can be reduced from O(Mn) to O(n). Applied to a real-world 32x102432 x 102432x1024 ultrasound dataset, Q-MERGE achieves an infidelity of 3.5x10−93.5 x 10^{-9}3.5x10−9, compared with 6.2x10−36.2 x 10^{-3}6.2x10−3 for direct preparation using the same underlying method, a more than six orders of magnitude improvement at equal CNOT circuit depth. We demonstrate its feasibility experimentally with a 53 qubit circuit comprised of 954 1-qubit gates and 478 2-qubit gates, on the Quantinuum System Model H2 trapped-ion quantum computer. The prepared state is validated using shadow-overlap tomography. Numerical analysis of Haar-random states indicates consistent success probability for up to M=107M=10^7M=107 segments, supporting the scalability of Q-MERGE for encoding massive classical datasets.
It remains unclear whether quantum machine learning (QML) has real advantages when dealing with practical and meaningful tasks. Encoding classical data into quantum states is one of the key steps in QML. Amplitude encoding has been widely used owing to its remarkable efficiency in encoding a number of 2n2^{n}2n classical data into nnn qubits simultaneously. However, the theoretical impact of amplitude encoding on QML has not been thoroughly investigated. In this work we prove that under some broad and typical data assumptions, the average of encoded quantum states via amplitude encoding tends to concentrate towards a specific state. This concentration phenomenon severely constrains the capability of quantum classifiers as it leads to a loss barrier phenomenon, namely, the loss function has a lower bound that cannot be improved by any optimization algorithm. In addition, via numerical simulations, we reveal a counterintuitive phenomenon of amplitude encoding: as the amount of training data increases, the training error may increase rather than decrease, leading to reduced decrease in prediction accuracy on new data. Our results highlight the limitations of amplitude encoding in QML and indicate that more efforts should be devoted to finding more efficient encoding strategies to unlock the full potential of QML.
Existence of quantum systems in superposition states gives an exponential advantage over classical system in both time and space. In this paper we will look at reconstructing real time classical data as a pure quantum state. Each classical data sample eg. each pixel location in an images is indexed by an\mathbf{n}nbit string representing one of the basis superposition states. The amplitude of the state which represents probability of measuring the state can be encoded as the normalized value of the sample at that location. A parameterized circuit[2] is used to prepare the quantum state as superposition of basis states from initial ground state. Each basis state corresponds to a sample classical data (eg. a pixel in case of an image). Quantum amplitude estimation based on Grovers algorithm[15] is used to estimate the amplitudes of each of these states. The parameters of the circuit is adjusted to minimize a loss function using classical gradient descent algorithm. The normalized amplitude at the output of our circuit is multiplied by a normalization factor that can also be tuned to represent the original data sample. This enables us to use the original real time data in our loss function computation. Section 1 of the paper gives a brief introduction. Section II covers the related works. Section III talks about amplitude encoding. Section IV describes quantum state preparation using supervised Learning. Section V describes implementation of the algorithms and finally Section VI gives the final conclusion of the research paper along with a result from the algorithm.