Data Augmentation for Neutron Spectrum Unfolding with Neural Networks

Round 1

Reviewer 1 Report

The paper "Data Augmentation for Neutron Spectrum Unfolding with Neural Networks" by McGreivy et al., presents three algorithms for the generation of neutron energy spectra. The results are clear, although one potential concern is the small data set available for validation purpose.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper investigates different algorithms to generate synthetic spectra to be used for neural networks training in support of spectra unfolding. particularly, for radiation protection applications. The article is well-written, relevant, and will be of great interests for the readers of the journals. I recommend to publish it after minor revisions and clarifications. Please provide a point-by-point response:
1. Table 1 and Figs. 1--2 should be moved after the sections in which are referenced. Same comment is valid for several other figures and tables in the article. Please correct it.

2. Line 107.  Further explanations are needed to justify the use of the Leaky ReLU as activation function over other possible choices.

3. As noticed by the authors in the paper, the MSE is not an optimal metric despite making easier the comparison with other methods in literature. I would present the RMSE in addition to the MSE in the tables even if it was not used as a metric to obtain the fitting parameters. This would allow to gain a better understanding of the algorithm performance when used to unfold local spectral features.

4. It could be interesting to use the ambient dose equivalent  H*(x) as a metric or to embed it in the loss function. This comment can be addressed in the future work sections.

5. Table 4. Substitute "*" with "\times"

6. Line 441. Missing reference "FRUIT spectrum generation algorithm (section ??)"

Author Response

Please see the attachment

Author Response File: Author Response.pdf