Title:
Learning with Adaptive Layer Activation in Spiking Neural Networks

Author:
Janis Zuters


=============================================
Overall rating:
   Accept: I will argue for this paper
Originality:
   Weak Accept: Vote accept, but won't object
Significance of topic:
   Neutral: Not impressed, won't object
Technical quality:
   Accept: I will argue for this paper
Relevance to Baltic DB&IS:
   Accept: I will argue for this paper
Presentation:
   Accept: I will argue for this paper
Referee's expertise on the topic
   Medium
Amount of rewriting required
   Medium
Optional: Which of the following session(s) would be the most appropriate for this paper?
   Information Systems and AI Technologies (Application)


--------------------------------------------------------------------------
Comments for the Program Committee (confidencial comments for PC use only)
--------------------------------------------------------------------------

None


---------------------------------------------------------------------------------------
Comments for the Authors (Constructive comments to the author(s) would be appreciated.)
---------------------------------------------------------------------------------------

   - Main contributions: The topic might not be of great interest of conference's but the experiment with its future expantion is interesting enough.

   - Positive aspects: The descriptions are clear so that an interested party could repeat the identical experiment.

   - Negative aspects: How did the auther determin those parameters is not clear. Some of the topic that the author neglect might be important -- Like Capacity dependence on the number of neurons.

   - Further comments: See below.




-----------------------------
As for expression in English:
-----------------------------

   - "server" might be "serve" 
           In "... Aacc is to server for it" 
   - "supervised and supervised" should be "supervised and unsupervised"
           In "Both supervised and supervised learning methods are available"
   - "simplification" should be " a simplification"
           In "Eq. (14) is simplification of Eq. (8)"
   - "A the moment threshold crossing" might be "At the moment of ..."
   - "Thus Eq. (1)" should be indented or included the previous paragraph.

(Anyway, proofread the manuscript more carefully!)

----------------------------------------------
Other minor recommendations as for its format:
----------------------------------------------
(1)
"1<=f<=n" would be better to be "f = 1, 2,..., n" in Eq. (2) since f is discrete.

(2)
format of "Rererence list" must be unified. E.g. 2002. or (2002).

(3)
Abstract should be independent. So, the acronim SNN should be given in the text not in the abstract.

(4)
"(wij)" in the caption of Fig. 1 should be "wij" comparing with the other such notations in the caption.

-------------------------
Recommendation to author:
-------------------------
(1)
Fig. 3 must be explained more in detail in its caption, as well as in the text.

(2)
The number of patterns recognized (capacity) might be dependent on number of neurons whichever just in one hidden-layer or those in the case of multiple layers. The author might refer to this point more explicitly even if it would be one of future works.

-----------------------------
Reviewer's opinions/questions
-----------------------------
(1)
Author wrote "... in order to avoid weights to become 'hollow' (too great or too small)." IMHO, however, it would be no need to avoid weight value becomes too small, that is, we may simply interpret the case the connection is pruned if such connections are not many.

(2)
One thing proposed in this paper is "adding acctivation accelaration factor Aacc," then the experimental results WITH AND WITHOUT this factor must be important for the sake of comparison.

(3)
How the author determined those parameters given in Table-1? After trial and error experiments? Or, Are those values are optimesed ones?