Review Reports - Spiking Neural Networks: History, Current Status and the Future

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The author gives a broad overview over Spiking Neural Networks. In fact, the historical review is really interesting. The paper gives some history of the development of neural models, and their use for modelling biological and cognitive phenomena, and for machine learning. It would be helpful if also soem considerations such as stability would be adressed. As the background of many readers will be from machine learning, a deeper consideration of ,cognitive phenomena would be inetresting. Also, to make the paper more interesting, some examples could be highlighted.

The paper also introduces the current state of the art in computational biological neuron and synapse modelling, and plasticity. again, for the engineering applications, some representative examples would be really interesting. Which are the highlights in applications in the last 20 years? And what is also a current issue is security, so how reproducible are results, how stable and can neural networks be unsed in critical applications (as for example energy managing networks, aerospace or autonomous driving)?

Spiking networks are also used for machine learning. As AI is one of the most discussed issues nowadays the author could add also some insights to specific AI algorithms based on these networks, autoencoders, adversorial networks etc.

The paper concludes with some project proposals that are all in all quite inetresting for discussing open questions and future work. Maybe there also some "prophetics" regarding what are we going to achieve in 10-30 years would be inetresting.

Author Response

Comments 1:

Response 1:

I added a para in the neurocognitive model section starting "Note that the
reproducibility of a spiking net depends in which it is implemented."

Comments 2:

As the background of many readers will be from machine learning, a deeper consideration of ,cognitive phenomena would be inetresting. Also, to make the paper more interesting, some examples could be highlighted.

Response 2:
I think the mentioned Markram et al paper is a highlight. I have added a new fourth paragraph in the biological models section starting "Biologists are interested".

Comments 3:

Which are the highlights in applications in the last 20 years? And what is also a current issue is security, so how reproducible are results, how stable and can neural networks be unsed in critical applications (as for example energy managing networks, aerospace or autonomous driving)?
I'm not sure what security means in this context. The para mentioned in the
prior response addresses reproducibility.

Response 3:

Added the penultimate paragraph in section 8.1 starting "It should be noted that both Hopfield nets and cell assemblies" to provide a link to autoencoders.

Comments 4:

Response 4:
I think there is already a good deal of future work. More prophetics, though I find it
interesting, is perhaps a bit much.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript “Spiking Neural Networks: History, Current Status and the Future” is a review article focusing on neural networks. I appreciate reading the manuscript. The subject of the review is interesting, and the manuscript is readable. In several parts, the manuscript adopts a colloquial tone, often including informal expressions alongside ambitious claims. I would suggest maintaining a more formal and consistent academic tone throughout. I suggest revision before recommending publication.

- In the caption of Figure 1, “almost no energy retained” appears to be incorrect and should read “almost no energy lost,” since a small leak implies minimal energy dissipation between time steps. Further, the colour lines are confusing; it is hard to distinguish between lines. I suggest changing colours and making the figure clearer.

- R65: The description of energy retention in Figure 1 seems inconsistent with the reported leak value. A small leak would typically imply minimal energy loss rather than minimal retention. Clarification would be appreciated.

- Figure 2 lacks clarity. The timeline resembles a tree structure, and the chronological progression is difficult to follow. It may be helpful to include brief keywords highlighting the main contribution of each work. I suggest revising the figure to improve readability.

- The manuscript contains a substantial number of acronyms. A consolidated table of abbreviations would enhance clarity and accessibility for readers.

- R177: The sentence appears grammatically incomplete, as the second clause lacks a main verb. The sentence should be: “There is not yet a good understanding of how neurons generate cognitive behaviour; however, connectionist systems including Hopfield networks [26], multi-layer perceptrons [27], and self-organising maps [28] have furthered our understanding of neural and connectionist computations.”.

-In Figure 3, it would be helpful to distinguish excitatory and inhibitory synapses using different colours or line styles. This would improve clarity and make the network topology easier to interpret.

- The manuscript refers to an “Iris categorisation system” without explicitly defining the dataset. It would be helpful to briefly specify that this refers to the Fisher Iris dataset and to include a short description of its structure.

- Figure 6 is difficult to interpret. The points are hard to distinguish. I suggest revising the figure to improve clarity.

Author Response

Comments 1: The manuscript Spiking Neural Networks: History, Current Status and the Future is a review article focusing on neural networks. I appreciate reading the manuscrnipt. The subject of the review is interesting, and the manuscript is readable. In several parts, the manuscript adopts a colloquial tone, often including informal expressions alongside ambitious claims. I would suggest maintaining a more formal and consistent academic tone throughout.

response 1:
Removed "I do most of my work".

comments 2:

I suggest revision before recommending publication.

- In the caption of Figure 1, "almost no energy retained" appears to be incorrect and should read "almost no energy lost," since a small leak implies minimal energy dissipation between time steps. Further, the colour lines are confusing; it is hard to distinguish between lines. I suggest changing colours and making the figure clearer.
- R65: The description of energy retention in Figure 1 seems inconsistent with the reported leak value. A small leak would typically imply minimal energy loss rather than minimal retention. Clarification would be appreciated.

Response 2:

The caption is correct. Line 1 is the, now, red line that stays near -65mv. I did change the colour of the lines. I agree that it makes the figure clearer. I also added a parenthetical statement to the text stating that it might actually be better called the retention constant, and a parenthetical (retention) to the caption. (Thanks)

Comments 2:
- Figure 2 lacks clarity. The timeline resembles a tree structure, and the chronological progression is difficult to follow. It may be helpful to include brief keywords highlighting the main contribution of each work. I suggest revising the figure to improve readability.

response 2:
I modified the figure so the refs slope and they are all on two (parallel) lines.

Comments 3:

The manuscript contains a substantial number of acronyms. A consolidated table of abbreviations would enhance clarity and accessibility for readers.

Response 3: There are 6, LIF, STDP, WTA, PID, IAT and CA. I don't think this warrants a table.

Comments 4:

- R177: The sentence appears grammatically incomplete, as the second clause lacks a main verb. The sentence should be: "There is not yet a good understanding of how neurons generate cognitive behaviour; however, connectionist systems including Hopfield networks [26], multi-layer perceptrons [27], and self-organising maps [28] have furthered our understanding of neural and connectionist computations.".

Response 4: further is the main verb, but I can see the confusion, so have changed it to have furthered. (Thanks)

Comments 5:

Made the excitatory plastic synapses dashed arrows and modified the caption.

Comments 6:

- The manuscript refers to an "Iris categorisation system" without explicitly defining the dataset. It would be helpful to briefly specify that this refers to the Fisher Iris dataset and to include a short description of its structure.

Response 6: I added the Fisher citation.

Reviewer 3 Report

Comments and Suggestions for Authors

Report on the paper: Spiking Neural Networks: History, Current Status and the Future, by C. Huyck.

In this review the author dicusses about spiking neural networks. The author gives a broad description about the historical development of the main models for spiking neural networks, giving some insights about the advantages and disadvantages of them, either in a physical context and biological meaning.

This kind of reviews are very valuable, pointing out in a critical way the main references about the existing models allows the reader to have a somehow broad panorama of the state of the art.

This paper is a review on the spiking nets, focused in four areas,
1. Simulation of biological neurons and networks.
2. Balanced Networks
3. Neurocognitive models
4. Machine Learning systems

I find the review well written but not truly well motivated, in the sense that there are already many reviews on the topic, actually many of them cited by the author himself. In the introduction it is not clearly stated what is the main motivation to have yet another review on spiking networks. In line 428 (p. 11), it is stated that the review [104] in the references of the paper, quote: “does not consider biology or psychology, unlike the paper you are currently reading”, which it is strange to me, since at least on the psychology part, I also do not find it within the paper.

I think another review on spiking networks is still interesting for the field, but it has to be well motivated, it has to have a clearly stated point of view, or new perspective which is discussed and contrasted to the existing literature. So, in my opinion his paper must be improved in that direction. Mostly because there are already many reviews on the topic which contains most of the parts considered in this paper and with more mathematical, biological or physical details than this one.

The conclusions of the paper are more like a closing thought. This is definitely because
the paper is a Review, which points out the advantages and disadvantages of previous works,
Considering even previous reviews on the topic.
The Chapter 8, of Discussion the author talks about his own interests on Parallel Processing and Neuromorphic computation as well as in AI, that is something that author can take advantage of, and try to discuss in more detail within the paper. What I find very valuable in the Discussion chapter, is the author’s perspective about future work.

The author seems to be aware of the references in the literature, it seems that the paper is a review of review, but even if this were the case, I feel an absence of deepness in the discussion of those reviews.
There are other reviews not cited by the author, and other that despite the fact they are cited they are mentioned without details, reviews like the one by Peiffer et al. (fnins), or Guo et al. (fnins) as well as Yamazaki et al. (brain science) are cited but in my opinion not very deeply discussed.

A review by Zhou et al in Frontiers in Neuroscience 2024, can be useful. Another very complete review by Majumdar in Evolving Systems 2025, would be useful to look at.

In conclusion, I would recommend this paper for publication after a revision, trying to explore in more detail the mathematical aspects of the major areas of focus of this paper.

Author Response

Comments 1: In this review the author dicusses about spiking neural networks. The author gives a broad description about the historical development of the main models for spiking neural networks, giving some insights about the advantages and disadvantages of them, either in a physical context and biological meaning.

This kind of reviews are very valuable, pointing out in a critical way the main references about the existing models allows the reader to have a somehow broad panorama of the state of the art.

This paper is a review on the spiking nets, focused in four areas,
1. Simulation of biological neurons and networks.
2. Balanced Networks
3. Neurocognitive models
4. Machine Learning systems

I find the review well written but not truly well motivated, in the sense that there are already many reviews on the topic, actually many of them cited by the author himself. In the introduction it is not clearly stated what is the main motivation to have yet another review on spiking networks.

Response 1:
I added the third paragraph in the introductory section starting "While there have been many recent review papers" I hope the motivation is now clear.

Comments 2:

In line 428 (p. 11), it is stated that the review [104] in the references of the paper, quote: "does not consider biology or psychology, unlike the paper you are currently reading", which it is strange to me, since at least on the psychology part, I also do not find it within the paper.
Response 2: I added pointers to the sections on biology and psychology into this sentence.

Comments 3: I think another review on spiking networks is still interesting for the field, but it has to be well motivated, it has to have a clearly stated point of view, or new perspective which is discussed and contrasted to the existing literature. So, in my opinion his paper must be improved in that direction. Mostly because there are already many reviews on the topic which contains most of the parts considered in this paper and with more mathematical, biological or physical details than this one.

Response 3: The added third introductory paragraph addresses this. I am unaware of a review paper on spiking nets that has more biological or physical details than this paper.

Comments 4:

The conclusions of the paper are more like a closing thought. This is definitely because
the paper is a Review, which points out the advantages and disadvantages of previous works,
Considering even previous reviews on the topic.
The Chapter 8, of Discussion the author talks about his own interests on Parallel Processing and Neuromorphic computation as well as in AI, that is something that author can take advantage of, and try to discuss in more detail within the paper. What I find very valuable in the Discussion chapter, is the author's perspective about future work.

The author seems to be aware of the references in the literature, it seems that the paper is a review of review, but even if this were the case, I feel an absence of deepness in the discussion of those reviews.
There are other reviews not cited by the author, and other that despite the fact they are cited they are mentioned without details, reviews like the one by Peiffer et al. (fnins), or Guo et al. (fnins) as well as Yamazaki et al. (brain science) are cited but in my opinion not very deeply discussed.

A review by Zhou et al in Frontiers in Neuroscience 2024, can be useful. Another very complete review by Majumdar in Evolving Systems 2025, would be useful to look at.

Response 4: Thanks for these. I put in a paragraph for Zhou at the end of the reinforcement learning section and one for Majumdar at the end of the control theory section.

Comments 5 : In conclusion, I would recommend this paper for publication after a revision, trying to explore in more detail the mathematical aspects of the major areas of focus of this paper.

Response 5: There is a lot of math one could explore, lots of neural models, plasticity models, balanced networks, and attractor dynamics. I think adding a little, or even a lot, would distract from the review as it sits. The math is of course described in the references.

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

After reading the author's responses and the new version of the paper, I consider the paper is ready. With the responses in mind, I agree that a deeper discussion on the mathematical aspects might distract the review from its main purpose.