SciPost Submission Page
Quantifying the Effects of Noise in a Quantum Convolutional Neural Network
by Zeyu Fan, Jonathan Wei Zhong Lau, Leong-Chuan Kwek
Submission summary
| Authors (as registered SciPost users): | Zeyu Fan |
| Submission information | |
|---|---|
| Preprint Link: | scipost_202407_00021v2 (pdf) |
| Date accepted: | Nov. 18, 2025 |
| Date submitted: | July 22, 2025, 11:57 a.m. |
| Submitted by: | Zeyu Fan |
| Submitted to: | SciPost Physics |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approaches: | Theoretical, Computational |
Abstract
This study quantifies the effects of quantum noise on the performance of a quantum convolutional neural network (QCNN), building on parallels with classical convolutional neural networks (CNNs), where added Gaussian noise can improve training speed, accuracy, and generalizability. While such benefits are established for classical CNNs, the influence of noise on quantum counterparts remains insufficiently characterized. We specifically examine three types of quantum noise: decoherence, Gaussian noise arising from imperfect quantum gates and experimental error, and systematic noise introduced during input state preparation. Our analysis provides a detailed assessment of how these distinct noise sources affect QCNN operation and outlines considerations for mitigating their impact. Though a QCNN is used as an example in this work, the methods used here can be applied to other quantum machine learning models as well.
Author indications on fulfilling journal expectations
- Provide a novel and synergetic link between different research areas.
- Open a new pathway in an existing or a new research direction, with clear potential for multi-pronged follow-up work
- Detail a groundbreaking theoretical/experimental/computational discovery
- Present a breakthrough on a previously-identified and long-standing research stumbling block
List of changes
-
Change in topic from finding advantages in introducing noise to the QCNN, to quantifying the effects of noise in the QCNN, as well as some proposals for mitigating their impacts.
-
The "Introduction" and "Datasets" sections have been expanded to include key details such as the inputs to the QCNN and how outputs are retrieved, which were previously only described in the "Methodology & Implementation" section. This is intended to make the main text more self-contained and easier to follow.
-
Modified Figures 10 and 18 for better clarity.
-
Reframed language throughout the Results, Discussion, and Conclusion sections to clarify that the focus is on quantifying the effects of noise on QCNN training and performance, rather than investigating potential benefits.
-
Removed speculative statements suggesting that noise may be beneficial in specific cases (e.g., MNIST with Depolarizing Channel noise) and replaced them with descriptions of observed effects without implying performance improvement.
-
Revised the Conclusion section to emphasize quantification of noise effects and identification of conditions where noise impacts training dynamics, while avoiding interpretations beyond what is statistically supported by the results.
-
Correction of numerous typos and errors. Apologies and thank you for catching them!
Current status:
Editorial decision:
For Journal SciPost Physics Core: Publish
(status: Editorial decision fixed and (if required) accepted by authors)
Reports on this Submission
Report #2 by Anonymous (Referee 3) on 2025-11-12 (Invited Report)
The referee discloses that the following generative AI tools have been used in the preparation of this report:
I used generative AI tools to improve the writing of the report
Report
The thorough examination of both a quantum phase recognition task and a classical image classification task provides an extensive viewpoint. I consider the data and analysis presented in the manuscript to be strong and a true addition to the field
However, I also see fundamental limitations that make me think that the manuscript is better suited for a journal like SciPost Physics Core. Indeed, the core finding is that quantum noise (Gaussian, Depolarizing Channel) is generally detrimental to training. While this should not be taken for granted (negative results are results and can be very valuable), I do not foresee an immediate impact of such exploratory work. Also, in the same direction, while the study is focused on how to quantify the problem of noise effect, the solutions offered are limited.
Recommendation
Accept in alternative Journal (see Report)
Report #1 by Anonymous (Referee 2) on 2025-10-8 (Invited Report)
The referee discloses that the following generative AI tools have been used in the preparation of this report:
report written by myself, then polished with ChatGPT 4o, then fixed by me again.
Report
Recommendation
Accept in alternative Journal (see Report)