Asynchronous Hierarchical Federated Learning Based on Bandwidth Allocation and Client Scheduling

Round 1

Reviewer 1 Report

Authors should personalize the evaluation of the clients as well. It would increase the impact of the presented study by understanding what is happening at the client level and how clients are affected by this aggregation.

In Figures 3 and 4, is the training time measured in seconds or minutes?

Please find the corrected mistakes below:

Abstract:

Line 3: "addresses" instead of "address" Line 7: Add a comma after "pre-allocation" Introduction:

Line 20: Change "facilitating" to "to facilitate" Line 24: Change "adheres to the principles" to "adheres to principles" Section Headings:

Section 2: Capitalize "Related Work" Section 3.1:

Subsection 3: Change "models" to "model" Section 3.3:

Line 274: Remove spaces around ellipses Section 3.4:

Line 284: Change "constraint" to "constraints" Line 302: Remove extra space in "attribute index" Section 4.1:

Line 356: Change "ReLu" to "ReLU" References:

Line 463: Remove extra space before period

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

abstract is not well-written,shall re-write
Introduction is too wordy. The flow of introduction is not clear to the readers
Related works are not promising to express the research gap. Need more literature and should elaborate what is done earlier and what need to be done.
All the references shall be cited as per the need.
In what way this reference is useful to this research Wen, L.; Li, X.; Gao, L.; Zhang, Y. A new convolutional neural network-based data-driven fault diagnosis method. IEEE 486 Transactions on Industrial Electronics 2017, 65, 5990–5998.
The fig.1, can be elaborated better. Why cache hierarchy is not mentioned in edge-cloud environment
Fig.2 is not understandable , why table should be a figure ?
Nothing is mentioned about simulation environment, how fig 1 is implemented
Recommended the authors to check the flow of contents in a order and it is of interest of the reader
Authors should address the need of the problem statement and compare with the trade-off methods
conclusion shall state the outcome.

Moderate, Once a run through  with the English language  professionals is recommended.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The asynchronous hierarchical FL technique presented here takes advantage of bandwidth pre-allocation. Authors should address the following remarks.

Authors should compare the proposed algorithm's computation time with the existing one. 

Abstract should highlight the novelty of the proposed work.

Also, the training time and energy consumption study has to be done in detail

Authors should correct the grammatical errors

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Authors have explained the queries clearly.

Reviewer 3 Report

All comments were addressed