Paper: 346955
Title: Kalman Gain Calculations with a Neural Network

-------------------- review 1 --------------------
---------------------------- REVIEW 1 --------------------------
PAPER: 6
TITLE: Kalman Gain Calculations with a Neural Network
 
OVERALL RATING: 2 (accept) 
REVIEWER'S CONFIDENCE: 3 (high) 
Relevance to this conference: 4 (good)
Originality/Uniqueness: 4 (good)
English readability: 4 (good)
Paper organization/presentation: 4 (good)
Has good survey been done?: 4 (good)


The paper is good. 

In my opinion, the principles of operation of a back-propagation neural network at matrix inversion should be described shortly. This operation is the key method of the paper, so it should be explained in some details.

Also, results of simulations should be explained more clearly.

Some technical recommendations are below.

1. References should be made in accordance with format requirements of the ICNNAI-2010.
2. Symbols in figures 1 and 2 are too small. It is difficult to read these notations.
3. Variables in the text should be made more accurately.

Final conclusion: the paper can be accepted after small improvement. 



-------------------- review 2 --------------------
---------------------------- REVIEW 2 --------------------------
PAPER: 6
TITLE: Kalman Gain Calculations with a Neural Network
 
OVERALL RATING: 2 (accept) 
REVIEWER'S CONFIDENCE: 4 (expert) 
Relevance to this conference: 5 (excellent)
Originality/Uniqueness: 4 (good)
English readability: 4 (good)
Paper organization/presentation: 4 (good)
Has good survey been done?: 1 (very poor)


The article presents an interesting way of gain ration calculation for the Kalman filter via neural network, eliminating a costly matrix inversion step influencing the stability of filters. 

The state of the art section is missing, thus it isn't clear if the proposed idea is an author's invention or he/she examine an already proposed solution still uncommon and hopes to popularise it. The revised paper should explain this issue.

As the original Kalman filter is a bit out-dated (due to linear characteristic) and could be exchanged with modern nonlinear filters e.g. unscented or particle filters, a question arises  - are there any future plans to use a similar approach to other filters e.g. UKF or CDKF? I hope the revised paper will explain this question.

The article should follow the available draft - the reference section does not use Chicago citation style. I also recommend to point-out the three sentences following "Some of them are: " (page 1) and to centre all the equations. 



-------------------- review 3 --------------------
---------------------------- REVIEW 3 --------------------------
PAPER: 6
TITLE: Kalman Gain Calculations with a Neural Network
 
OVERALL RATING: 2 (accept) 
REVIEWER'S CONFIDENCE: 2 (medium) 
Relevance to this conference: 4 (good)
Originality/Uniqueness: 3 (fair)
English readability: 4 (good)
Paper organization/presentation: 4 (good)
Has good survey been done?: 3 (fair)


The paper presents an approach to replace a part of the Kalman gain calculations with a
neural network. The work can be useful, although does not seem to be too difficult in the 
presence of other work on matrix manipulations with neural networks.