RECORD
Drastic Circuit Depth Reductions with Preserved Adversarial Robustness by Approximate Encoding for Quantum Machine Learning
Quantum Physics · Submitted 18 Sep 2023
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim Our results, based on simulations and experiments on IBM quantum devices, highlight a promising pathway for the future implementation of accurate and robust QML models on complex datasets relevant for practical applications, bringing the possibility of NISQ-era QML advantage closer to reality.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Quantum machine learning (QML) is emerging as an application of quantum computing with the potential to deliver quantum advantage, but its realisation for practical applications remains impeded by challenges. Amongst those, a key barrier is the computationally expensive task of encoding classical data into a quantum state, which could erase any prospective speed-ups over classical algorithms. In this work, we implement methods for the efficient preparation of quantum states representing encoded image data using variational, genetic and matrix product state based algorithms. Our results show that these methods can approximately prepare states to a level suitable for QML using circuits two orders of magnitude shallower than a standard state preparation implementation, obtaining drastic savings in circuit depth and gate count without unduly sacrificing classification accuracy. Additionally, the QML models trained and evaluated on approximately encoded data display an increased robustness to adversarially generated input data perturbations. This partial alleviation of adversarial vulnerability, possible due to the "drowning out" of adversarial perturbations while retaining the meaningful large-scale features of the data, constitutes a considerable benefit for approximate state preparation in addition to lessening the requirements of the quantum hardware. Our results, based on simulations and experiments on IBM quantum devices, highlight a promising pathway for the future implementation of accurate and robust QML models on complex datasets relevant for practical applications, bringing the possibility of NISQ-era QML advantage closer to reality.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Quantum machine learning (QML) is emerging as an application of quantum computing with the potential to deliver quantum advantage, but its realisation for practical applications remains impeded by challenges. | background 1.00 | 0.04 | 0.02 | asserted 0.64 |
| s1 | Amongst those, a key barrier is the computationally expensive task of encoding classical data into a quantum state, which could erase any prospective speed-ups over classical algorithms. | background 0.97 | 0.05 | 0.03 | hedged possibility 0.76 |
| s2 | In this work, we implement methods for the efficient preparation of quantum states representing encoded image data using variational, genetic and matrix product state based algorithms. | method 0.67 | 0.94 | 0.04 | asserted 0.99 |
| s3 | Our results show that these methods can approximately prepare states to a level suitable for QML using circuits two orders of magnitude shallower than a standard state preparation implementation, obtaining drastic savings in circuit depth and gate count without unduly sacrificing classification accuracy. | result 1.00 | 0.97 | 0.62 | asserted 0.99 |
| s4 | Additionally, the QML models trained and evaluated on approximately encoded data display an increased robustness to adversarially generated input data perturbations. | result 1.00 | 0.94 | 0.15 | asserted 1.00 |
| s5 | This partial alleviation of adversarial vulnerability, possible due to the "drowning out" of adversarial perturbations while retaining the meaningful large-scale features of the data, constitutes a considerable benefit for approximate state preparation in addition to lessening the requirements of the quantum hardware. | interpretation 0.99 | 0.82 | 0.16 | asserted 0.68 |
| s6 | Our results, based on simulations and experiments on IBM quantum devices, highlight a promising pathway for the future implementation of accurate and robust QML models on complex datasets relevant for practical applications, bringing the possibility of NISQ-era QML advantage closer to reality. | interpretation 0.85 | 0.94 | 0.65 | hedged possibility 0.54 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Quantum machine learning (QML) is emerging as an application of quantum computing with the potential to deliver quantum advantage, but its realisation for practical applications remains impeded by challenges. Amongst those, a key barrier is the computationally expensive task of encoding classical data into a quantum state, which could erase any prospective speed-ups over classical algorithms. In this work, we implement methods for the efficient preparation of quantum states representing encoded image data using variational, genetic and matrix product state based algorithms. Our results show that these methods can approximately prepare states to a level suitable for QML using circuits two orders of magnitude shallower than a standard state preparation implementation, obtaining drastic savings in circuit depth and gate count without unduly sacrificing classification accuracy. Additionally, the QML models trained and evaluated on approximately encoded data display an increased robustness to adversarially generated input data perturbations. This partial alleviation of adversarial vulnerability, possible due to the "drowning out" of adversarial perturbations while retaining the meaningful large-scale features of the data, constitutes a considerable benefit for approximate state preparation in addition to lessening the requirements of the quantum hardware. Our results, based on simulations and experiments on IBM quantum devices, highlight a promising pathway for the future implementation of accurate and robust QML models on complex datasets relevant for practical applications, bringing the possibility of NISQ-era QML advantage closer to reality.
Human review
How the literature uses it
7 arXiv papers cite this in 9 sentences
Papers citing it · 5 sentences without a confident label
Cites it as support 2 papers
-
The number of gates required to produce the approximate amplitudes is far smaller than the number required to produce the exact amplitudes (see, for example, [ref] ).
Radio Signal Classification by Adversarially Robust Quantum Machine Learning · Approximate Amplitude Encoding · 0.97
-
This follows from Ref. [ref] where the authors found that training QNNs with noisy inputs results in a notable improvement in test accuracy when tested on similarly noisy inputs.
Calibrating the role of entanglement in variational quantum circuits · Quantum Machine Learning · 0.99
Cites it as background 2 papers
-
However, evolutionary strategies have also been proposed and applied to certain tasks, such as finding molecular or many-body ground states [ref] , quantum encoding and state preparation [ref] , as well as classically defined tasks [ref] .
Quantum Time-Series Learning with Evolutionary Algorithms · Introduction · 0.87
-
There are several previous methods that attempt to address the input problem [ref] ; we highlight some prominent approaches here.
Data is often loadable in short depth: Quantum circuits from tensor networks for finance, images, fluids, and proteins · Introduction · 0.96
What it cites · 19 sentences without a confident label
Uses the method or tool of 3 papers
-
Motivated by the notion that accurate QML models can be implemented on very noisy encoded states we implement three methods of approximate state preparation and benchmark them by both their raw accuracy and adversarial robustness: a deterministic technique for constructing a matrix product state (MPS) approximation to the target state [ref] , and two machine learning based methods which attempt to learn unitaries that perform the state preparation, relying in one case on genetic algorithms (GASP) [ref] and in the other on variational quantum circuits.
Efficient quantum state tomography · Introduction · 0.94
-
Initially $𝒱(θ)$ is composed of a single such layer, and its parameters are optimised with 100 steps of the Adam optimiser [ref] .
Adam: A Method for Stochastic Optimization · Introduction · 1.00
-
We generate adversarial perturbations by the method of projected gradient descent (PGD) [ref] , which amounts to attempting to solve the optimisation problem of Equation [ref] iteratively through a slight modification of standard gradient descent.
Towards Deep Learning Models Resistant to Adversarial Attacks · Introduction · 1.00
Uses data or values from 1 paper
-
Approximate State Preparation Our investigation focuses on the problem of image classification, with images drawn from the MNIST [ref] and FMNIST [ref] datasets (as well as, later, a simple dataset constructed for the purpose of running small QML models on real devices).
Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms · Introduction · 0.94
Cites as background 11 papers
-
Matrix Product State Encoding: Matrix Product States (MPS) are a tensor network representation of quantum systems that have found significant application in the simulation of quantum algorithms as a result of their ability to represent and manipulate states of relatively low bipartite entanglement in a memory efficient manner [ref] .
Efficient classical simulation of slightly entangled quantum computations · Introduction · 0.85
-
Concurrently, the emergence of programmable quantum computers has led to intense interest in the prospect of quantum machine learning (QML) [ref] – the study of ML algorithms which exploit the capabilities of quantum computers.
Hybrid Quantum-Classical Generative Adversarial Network for High Resolution Image Generation · Introduction · 0.99
-
Concurrently, the emergence of programmable quantum computers has led to intense interest in the prospect of quantum machine learning (QML) [ref] – the study of ML algorithms which exploit the capabilities of quantum computers.
Supervised quantum machine learning models are kernel methods · Introduction · 0.99
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.