Ultrafast Laser Beam Profile Characterization in the Front-End of the ELI-NP Laser System Using Image Features and Machine Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper analyzes the stability of spatial-energy parameters of femtosecond laser radiation.

The paper should indicate standard methods for measuring the parameters of femtosecond lasers (for example, ISO 11146).

Figure 2 shows the laser beam spots. The transverse intensity distribution of one of the implementations should be shown to understand its type (Gaussian distribution or other).

It is also desirable to show the operation of the described algorithm for an unstable laser (it is enough for several parameters).

Formulas (1) and (7) are not displayed correctly in the manuscript.

Author Response

Response to Reviewer 1

We are grateful to the reviewer for his constructive and insightful comments on our manuscript titled "Ultrafast Laser Beam Profile Characterization in the Front-End of the ELI-NP Laser System Using Image Features and Machine Learning." Below we address each point raised, with justifications reflected in the revised manuscript (highlighted).

 

Reviewer Comment 1:

The paper should indicate standard methods for measuring the parameters of femtosecond lasers (for example, ISO 11146).

Response:
We appreciate this important suggestion. In response, we have referenced ISO 11146 in both the Introduction and Discussion sections of the manuscript. We acknowledge its role in providing standardized spatial beam parameter measurements (e.g., M², waist, divergence). Our work is framed as a complementary approach that leverages image-based diagnostics and machine learning to capture fine-scale instabilities and spatial deviations not directly covered under ISO methods.

 

Reviewer Comment 2:

Figure 2 shows the laser beam spots. The transverse intensity distribution of one of the implementations should be shown to understand its type (Gaussian distribution or other).

Response:
Thank you for this valuable observation. In the Beam Profile Analysis section, we now clarify that Frame 10 was visually assessed and found to exhibit a near-Gaussian transverse intensity distribution. Due to current image processing constraints, we did not perform formal 1D Gaussian fitting or extract intensity slices. Nevertheless, the intensity's qualitative shape and smooth falloff suggest a distribution consistent with typical femtosecond laser beam behavior. This addition helps contextualize the beam profile despite the absence of numerical fitting routines.

 

Reviewer Comment 3:

It is also desirable to show the operation of the described algorithm for an unstable laser (it is enough for several parameters).

Response:
We acknowledge the value of evaluating the algorithm under unstable conditions. However, we could not introduce instability experimentally due to safety protocols and the strict operational stability required by the ELI-NP laser system. As an alternative, we highlight in the Anomaly Detection section that natural variations in the dataset (notably Frame 10) already exhibit mild fluctuations in centroid, intensity, and edge sharpness. These serve as valid, real-world indicators of the algorithm's sensitivity to deviations without artificially generating synthetic data.

 

Reviewer Comment 4:

Formulas (1) and (7) are not displayed correctly in the manuscript.

Response:
Thank you for pointing this out. We have corrected the display of Equations (1) and (7), and now these equations appear clearly in the manuscript with appropriate mathematical structure.

 

Again, we thank the reviewer for the helpful feedback. These comments have helped us refine the presentation and clarity of our work without altering its original scope or objectives.

Sincerely

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript, Tayyab Imran et al “Ultrafast Laser Beam Profile Characterization in the Front-End of the ELI-NP Laser System Using Image Features and Machine Learning”, constructs a comprehensive framework for analyzing the ELI-NP Front-End beam characteristics. The authors combine the traditional beam diagnostics with machine learning techniques to monitor energy instabilities and resultant geometric parameter drifts, such as ellipticity, roundness, symmetries, etc. These anomalies are identified using an unsupervised isolation forest algorithm. Overall, the paper is very well-written. I am certain that the manuscript will become suitable for publication after my questions and concerns are swiftly addressed.

 

1. I appreciate the authors’ efforts to attempt to first understand the dataset with statistical observables. Can the authors append this analysis with a statistical analysis (for example, a \Chi-square test) on whether these outliers can be sifted out via standard statistics?
2. Following the previous question, can the authors benchmark the speed of statistical analysis vs the unsupervised machine learning analysis? Is machine learning fast enough for actual monitoring?
3. Is there an overfitting issue in this problem, given the rather small number of images? What is the tree size in this isolation forest?

Comments on the Quality of English Language

Good

Author Response

Response to Reviewer 2

We are grateful to the reviewer for his constructive and insightful comments on our manuscript titled "Ultrafast Laser Beam Profile Characterization in the Front-End of the ELI-NP Laser System Using Image Features and Machine Learning." Below we address each point raised, with justifications reflected in the revised manuscript (green highlighted).

 

Reviewer Comment 1:

Can the authors append this analysis with a statistical analysis (for example, a Chi-square test) on whether these outliers can be sifted out via standard statistics?

Response:
We sincerely thank the reviewer for the valuable suggestion regarding statistical testing, such as the Chi-square test, to strengthen outlier identification. We fully agree that incorporating formal statistical methods could enhance the robustness of anomaly detection. However, given the relatively small size of the dataset and the exploratory nature of this initial study, we have focused on outlier detection based on direct image features and unsupervised learning techniques.

In response to the reviewer's insightful comment, we have now acknowledged this limitation and discussed the potential integration of statistical hypothesis testing as a direction for future work in the revised Discussion section. We are grateful for this constructive feedback, which will help guide future developments in this research.

 

Reviewer Comment 2:

Can the authors benchmark the speed of statistical analysis vs the unsupervised machine learning analysis? Is machine learning fast enough for actual monitoring?

Response:
We appreciate this practical question. While our current implementation involves offline analysis, we have clarified in the Discussion section that the Isolation Forest algorithm processes each beam frame in under 3 milliseconds on standard hardware. This highlights the method’s computational efficiency and indicates its suitability for future integration into fast diagnostic workflows. Although we did not benchmark it directly against statistical methods, all techniques used were lightweight and efficient in processing the image data.

 

Reviewer Comment 3:

Is there an overfitting issue in this problem, given the rather small number of images? What is the tree size in this isolation forest?

Response:
Thank you for raising this concern. The Discussion now includes a detailed description of the Isolation Forest parameters: 100 trees with a subsample size of 64 per tree. These settings were chosen specifically to promote generalization and minimize the risk of overfitting. Additionally, since the algorithm is unsupervised and relies on randomized subsampling, the risk of overfitting is inherently reduced in this context.

 

We again thank the reviewer for their constructive feedback, which helped us enhance the clarity and robustness of our manuscript.

Sincerely