Research on Generalized Hybrid Probability Convolutional Neural Network

Round 1

Reviewer 1 Report

The manuscript discusses the generalized ability of the Convolution layers of CNN. A generalized hybrid probability convolutional neural network (GHP-CNN) is proposed to solve abstract feature classification of a distributed form. The authors have not discussed the Graph Convolution Network at all. Actually, the Graph CNN concept was brought to bring the generalized ability to tackle different problems. 

Some recent Graph CNN interesting papers are as follows:

Rama-Maneiro, E., Vidal, J. C., & Lama, M. (2021). Embedding Graph Convolutional Networks in Recurrent Neural Networks for Predictive Monitoring. arXiv preprint arXiv:2112.09641.

Nassar, M., Wang, X., & Tumer, E. (2019). Fully Convolutional Graph Neural Networks using Bipartite Graph Convolutions.

Hong, Y., Liu, Y., Yang, S., Zhang, K., Wen, A., & Hu, J. (2020). Improving graph convolutional networks based on relation-aware attention for end-to-end relation extraction. IEEE Access, 8, 51315-51323.

# What do you mean by the generalization ability of the convolutional layer to extract image features? How does the number of CNN layers, filtering features, stride, padding, and pooling affect this generalized ability of convolution layers?

# Mathematical complexity concerning the generalization ability must be discussed thoroughly.

# What about graph-based CNN? Comparing the convolutional layer's generalization ability with Graph-based CNN is recommended.

# What are the termination criteria used for learning?  

Author Response

Please see the attachment for the reply.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper deals with the generalization ability of the convolutional layer as a feature mapper to extract image features and the classification ability of the Multi-Layer Perception(MLP) in the CNN. A generalized hybrid probability convolutional neural network (GHP-CNN) is proposed to solve abstract feature classification with unknown distribution. The paper is more or less well written and results are correct. However, I have some concerns and the paper should be revised in accordance with the following:

1. Figure 2: a sudden spike (consistent in all the graphs) appeared at 50. Any explanation?

2. Equation (2): I do not see the whole point. On one hand, it is a well understood function. On the other hand- how is this connected to the present problem? 

3. Any comparison with respect to existing models in literature? In particular, are Table 4, 5 etc. improving existing results significantly?

 4. Data science based techniques (in particular CNN) is recently used in the following open access paper: A novel implementation of Siamese type neural networks in predicting rare fluctuations in financial time series, Risks, Vol. 10, No. 2: 39, 2022 (16 pages).

The authors should check the paper (and reference therein).

 

Author Response

Please see the attachment for the reply.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

 It is challenging to realize the modification in the revised version. It will be easy to recognize If the modifications are highlighted or shown in colored text.  Also, a response sheet can also serve the purpose.  

Author Response

Please see the attachment.

Author Response File: Author Response.docx