Reinforcement Learning-Based Augmentation of Data Collection for Bayesian Optimization Towards Radiation Survey and Source Localization
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsInteresting paper with some sound methodology. But it seems moderately incremental and makes much use of a particular reference on which the work seems to build. The ma8n weakness in the paper is that the conclusions seem not very conclusive. Especially for a real world scenario, one would expect a review of potential gains by a more complex algorithm (which takes significantly longer to compute) versus the survey speed and potential dose to the machine.
Author Response
Thank you for your criticism and time,
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe title is correct, although it takes long reading to link the article content to the title.
Somehow one gets the feeling of a starting hypothesis that in the end does not conclude, is the Reinforcement Learning improving BO or not. It takes definitely longer time and needs more start inputs.
The article gives an exhaustive introduction with references to parallel theoretical work performed by other groups. Thereafter is introduced the technique of RL-BO in a rather lengthy, very detailed in equations but not very explicatory.
Then comes the experimental part. The experimental procedure is not explained at all. Robots are being mentioned in the text further up but nothing here. What kind of robot what kind of detectors is not mentioned.
The figures and tables are not explained, what information should the reader appreciate what is better. How is Figure 3 related with the active source and Figure 2? A drawing of the overall layout of the building and source position would probably make it easier to take conclusions from figure
The computer used is mentioned (standard PC not very powerful) but not the software e.g. what operating system is it running.
The connection between theory, experiment and result is not well made.
The text as such is well written, the english is good.
Author Response
Thank you for your criticism and time,
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors have more or less responded to all my comments.
The added text makes it now more easy to appreciate the figures.
I beleive the article now is OK to be published.