2 \documentclass{article}
3 \usepackage[T1]{fontenc}
4 \usepackage[latin1]{inputenc}
7 \renewcommand{\labelenumii}{\labelenumi\arabic{enumii}}
15 \vspace{-0.5cm}\hspace{-2cm}Computer Science Laboratory, LIFC
17 \hspace{-2cm}University of Franche-Comt\'e
19 \hspace{-2cm}25030 Besan\c{c}on, France.
24 Detailed changes and addressed issues in the revision of the paper
26 "Recurrent Neural Networks and Chaos: Construction, \\
27 Evaluation, and Prediction Ability" \\
29 Neural Networks and Chaos: Construction, Evaluation of Chaotic Networks, \\
30 and Prediction of Chaos with Multilayer Feedforward Networks"
32 by Jacques M. Bahi, Jean-Fran\c{c}ois Couchot, \\
33 Christophe Guyeux, and Michel Salomon
38 Please, find below the detailed changes and issues we addressed in the
39 revision of our above mentioned paper that we resubmit.
41 All the remarks and recommendations of the three reviewers have been
42 considered and have led to modifications in the paper.
45 \item Concerning the remarks of reviewer No. 1:
49 \item Please explain the topic "Recurrent Neural Networks and Chaos:
50 Construction, Evaluation, and Prediction Ability" which makes me
51 confused. Does the paper discuss the construction, evaluation, and
52 prediction ability of chaotic neural networks? And in the paper, the
53 chaotic neural networks are mainly done research on.
57 Indeed, our contributions are twofold. Firstly, we present how to
58 construct chaotic neural networks, according to Devaney's definition
59 of chaos, and we discuss their evaluation. Secondly, we study the
60 capacity of classical multilayer feedfoward neural networks to
61 predict chaotic data. To highlight clearly these contributions we
62 have changed the title of our paper, as said above.
63 \item Please explain the Line 242( "than chaotic iterations Ff with
64 initial condition............" ) and Line 298( "investigate, when
65 comparing neural networks and Devaney's chaos").
70 \item Section VI analyzes the suitability of artificial neural
71 networks for predicting chaotic behaviors. So, I think section VI
72 don't correspond to the topic. And homoplastically, lines 30-32("the
73 learning, with neural networks having a feedforward structure, of
74 chaotic behaviors represented by data sets obtained from chaotic
75 maps, is far more difficult than non chaotic behaviors") refer to
76 the learning of neural networks having a feedforward structure. But
77 lines 402-405 refer to all the network topologies.
81 We have corrected lines 402-405 (now lines 419-422) as follows:
82 ``For all those feedforward network topologies and all outputs the
83 obtained results for the non-chaotic case outperform the chaotic
84 ones. Finally, the rates for the strategies show that the different
85 feedforward networks are unable to learn them.'' More generally, in
86 the whole paper the term feedforward has been added, when needed, to
87 clarify our discussion.
88 \item Please explain the meanings of the percentage in TABLE I and
93 We have explained how the success rates (percentage) are computed
95 \item Except for prediction success rates, in order to reflect the
96 prediction ability, please add the analysis of the prediction errors
97 for data sequence and diagram it.
101 We have added two figures (Figures~3 and 4) and some comments about them
102 lines~425-431. We hope that we have answered the reviewer remark.
106 \item Concerning the remarks of reviewer No. 2:
110 \item On the basis of the result botained by Guyeux in Ref[12], the
111 authors dealed with chaotic neural networks for various fields of
112 appli1cation. Firstly, the authors described how to build a neural
113 network that can be trained to learn a given chaotic map function,
114 then found a condition that allow to check whether the iterations
115 induced by a function are chaotic or not, and thus if a chaotic map
116 is obtained. As the authors said that this work is different from
117 most of prviews works, this manuscript gave a rigorous mathematical
118 proof for chaos of chaotic neural networks. This is a very
119 interesting work. On the other hand, I think, the conclusion is too
120 long, and some Definitions such as Definition 1-Definition 5 are
121 wellknow Definition, it's no necessary to be presented any more.
125 As suggested by the reviewer the conclusion has been shortened and we
126 have presented those definitions differently.
129 We are very grateful to the reviewers who, by their recommendations,
130 allowed us to improve our paper.
133 %\TeXButton{Flushright}{\begin{flushright}