Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper introduces an intriguing and well-designed approach to utilizing spiking neurons for programming tasks, particularly with the implementation of cell assemblies (CAs). The authors have done an excellent job outlining the core concept of binary cell assemblies and demonstrating how they can be used to implement complex tasks like finite state automata (FSA), cognitive mapping, and parsing. The ability to simulate and develop these systems with relatively simple neural models opens up exciting opportunities for neuromorphic computing and cognitive agent design. The inclusion of real-world applications like embodied agents in virtual environments is a commendable aspect.

2. Code Availability and Accessibility:

The paper mentions that the code for this research can be found at the following link:
https://cwa.mdx.ac.uk/chris/NEAL/simpleProgramming.html
However, I encountered difficulty accessing the provided link. The page appears to be broken, and I was unable to retrieve the code or examine the programming framework in greater detail. It is crucial that the authors recheck the link to ensure it is functional, as access to the code is an important part of evaluating the practical implementation of their proposed method.

3. Methodology and Results Section:

The overall organization of the paper's methodology section can be improved to provide better clarity and flow. Specifically, while the authors describe the various components such as the cell assemblies, parsers, and planners, I suggest they could benefit from integrating more of the experimental results and findings within the methodology section itself. This would make it easier for readers to directly correlate the design choices with the outcomes and effectiveness of these components. Including more detailed results or performance data in this section could help substantiate the claims made regarding the system's capabilities, particularly in relation to the practical use of spiking neural networks in real-world applications.

4. Additional Considerations:

• It may be beneficial to include a deeper discussion on the limitations or challenges associated with scaling up the proposed system. For instance, how does the system perform with larger sets of neurons or more complex environments?

• The paper might also benefit from further elaboration on the biological plausibility of the spiking neural network models used and how these models compare to real neural activity in the brain. Although the authors acknowledge that some aspects are not fully biologically accurate, a more detailed discussion of this would add depth to the paper’s contribution.

Conclusion:

This is a well-written paper that introduces a novel approach to programming using spiking neurons. The concept of cell assemblies and their application to various cognitive and computational tasks is compelling. With the accessibility of the code and a few adjustments to the methodology and results sections, this paper could become a significant contribution to the field of neuromorphic computing.

Comments on the Quality of English Language

Good.

Author Response

Thanks for this review.  I found it helpful and I think I addressed all of the concerns.  Below, my responses are prefixed with #, responding to the portion just above.

 

The paper introduces an intriguing and well-designed approach to utilizing spiking neurons for programming tasks, particularly with the implementation of cell assemblies (CAs). The authors have done an excellent job outlining the core concept of binary cell assemblies and demonstrating how they can be used to implement complex tasks like finite state automata (FSA), cognitive mapping, and parsing. The ability to simulate and develop these systems with relatively simple neural models opens up exciting opportunities for neuromorphic computing and cognitive agent design. The inclusion of real-world applications like embodied agents in virtual environments is a commendable aspect.

2. Code Availability and Accessibility:

The paper mentions that the code for this research can be found at the following link:
https://cwa.mdx.ac.uk/chris/NEAL/simpleProgramming.html
However, I encountered difficulty accessing the provided link. The page appears to be broken, and I was unable to retrieve the code or examine the programming framework in greater detail. It is crucial that the authors recheck the link to ensure it is functional, as access to the code is an important part of evaluating the practical implementation of their proposed method.
#Thanks for this.  It was a silly mistake.  On page 3 it was correct.
#https://cwa.mdx.ac.uk/NEAL/simpleProgramming.html.  It's now correct in the
#data availability statement too.

3. Methodology and Results Section:

The overall organization of the paper's methodology section can be improved to provide better clarity and flow. Specifically, while the authors describe the various components such as the cell assemblies, parsers, and planners, I suggest they could benefit from integrating more of the experimental results and findings within the methodology section itself. This would make it easier for readers to directly correlate the design choices with the outcomes and effectiveness of these components. Including more detailed results or performance data in this section could help substantiate the claims made regarding the system's capabilities, particularly in relation to the practical use of spiking neural networks in real-world applications.
#There isn't a methodology section, but I think the reviewer is referring to
#section 5 on the FSA and timers.  Reviewer 3 has suggested expanding the
#discussion of figures 2 and 3, so all of the components are described
#in the results subsections.  I added a paragraph in section 5 on the parser
#and one for the planner.

4. Additional Considerations:

It may be beneficial to include a deeper discussion on the limitations or challenges associated with scaling up the proposed system. For instance, how does the system perform with larger sets of neurons or more complex environments?
#A paragraph on the effect of adding neurons is added to the discussion.
#It's the para starting "the binary CA model is not".

The paper might also benefit from further elaboration on the biological plausibility of the spiking neural network models used and how these models compare to real neural activity in the brain. Although the authors acknowledge that some aspects are not fully biologically accurate, a more detailed discussion of this would add depth to the paper’s contribution.
#A paragraph on biological relevance is also added to the discussion.
#It's also the para starting "The binary CA model is not"

Conclusion:

This is a well-written paper that introduces a novel approach to programming using spiking neurons. The concept of cell assemblies and their application to various cognitive and computational tasks is compelling. With the accessibility of the code and a few adjustments to the methodology and results sections, this paper could become a significant contribution to the field of neuromorphic computing.

 

#Thanks again for the review.

 

Reviewer 2 Report

Comments and Suggestions for Authors

The authors present a framework that demonstrates how spiking neurons and binary cell assemblies can be used to build programmable, biologically inspired systems. I have a few questions regarding their approach and future directions, which need to be addressed:

1. How well does the binary cell assembly model scale when building larger and more complex neural systems while maintaining stability and biological relevance?
2. What are the main differences in performance or behavior when running NEAL on simulation platforms like Nest versus neuromorphic hardware such as SpiNNaker?
3. Are there plans to extend the NEAL framework beyond STDP-based binding to include more general learning mechanisms such as reinforcement learning or structural plasticity? Authors should discuss how incorporating these mechanisms might impact the system’s adaptability and biological relevance.
4. How could the CABot4 architecture be extended or adapted for real-world robotic or embodied AI applications?

Author Response

Thanks for this review.  I tried to answer the questions in the paper.  I've left the original reviewers response as is, and prefixed my responses with #.

 

The authors present a framework that demonstrates how spiking neurons and binary cell assemblies can be used to build programmable, biologically inspired systems. I have a few questions regarding their approach and future directions, which need to be addressed:

How well does the binary cell assembly model scale when building larger and more complex neural systems while maintaining stability and biological relevance?
#Using many binary CAs is discussed in a para in the discussion subsection.
#The para starting the binary CA model can be
#A paragraph on biological relevance is also added to the discussion.  
#The para starting the binary CA model is not 

What are the main differences in performance or behavior when running NEAL on simulation platforms like Nest versus neuromorphic hardware such as SpiNNaker?
#This is addressed in the same new paragraph as the wider use of the binary CAs.

Are there plans to extend the NEAL framework beyond STDP-based binding to include more general learning mechanisms such as reinforcement learning or structural plasticity?
#A reference to our existing work on reinforcement learning with CAs and Hebbian
#learning has been added to the future work.
#A sentence on structural plasticity is added to the paragraph starting
#Developmenatl psychological issues in the future work subsection.
Authors should discuss how incorporating these mechanisms might impact the system’s adaptability and biological relevance.
How could the CABot4 architecture be extended or adapted for real-world robotic or embodied AI applications?
#Robots were already mentioned in the future work subsection in the
#paragraph starting The approach used.  A new final sentence has been
#added to that paragraph.  

-Chris

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript can be accepted after addressing following concerns:

1. Abstract is less informative, it is hard to understand the work done in the paper based on abstract, make it more informative.
2. Line 74- “C_M is the membrane capacity”, make sure C_M is capacity or capacitance.
3. Elaborate the analysis of Figure 2 and 3.
4. “Discussion, Future work and conclusion” make this section concise.

Author Response

Thanks for this.  I've tried to address all of the concerns, but the last concern on conciseness of the last section contradicts the other two reviewers that wanted more in that section (or at least that I put in that section).  I've left reviewer 3's comments as they came, and my responses are prefixed with #.

 

This manuscript can be accepted after addressing following concerns:

Abstract is less informative, it is hard to understand the work done in the paper based on abstract, make it more informative.
#The abstract has been extended.

Line 74- “CM is the membrane capacity”, make sure CM is capacity or capacitance.
#Yes, of course it's capacitance.  Changed.  Thanks.

Elaborate the analysis of Figure 2 and 3.
#Added two paragraphs in the Simple CABot4 Results subsection for
#figure 2.  Added a paragraph and a few sentences to existing paragraphs
#in section CABot4 results for figure 3.  

“Discussion, Future work and conclusion” make this section concise.
#We have broken this section into two, with a concise conclusion,
#and a section with two subsections on Discussion and Future work.
#We expect this is not what reviewer 3 intended, but the other
#reviewers requested more discussion and future work.  

 

Thanks again.  I hope this addresses the reviewer's concerns.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thanks for the responses.  

Since the paper focuses on developing the programming with Spiking Neurons, I recommend discussing more recent papers about spiking neural networks such as: 

Ana Stanojevic, et al. High-performance deep spiking neural networks with 0.3 spikes per neuron. 2024 Nature Comm. 

These above studies employ different kinds of spiking neural network models, and please briefly discuss them about how these models are applied to the proposed programming tools.

Comments on the Quality of English Language

Good.

Author Response

Reviewer 1

Since the paper focuses on developing the programming with Spiking Neurons, I recommend discussing more recent papers about spiking neural networks such as: 

Ana Stanojevic, et al. High-performance deep spiking neural networks with 0.3 spikes per neuron. 2024 Nature Comm. 

These above studies employ different kinds of spiking neural network models, and please briefly discuss them about how these models are applied to the proposed programming tools.

Response

Thanks for this.  It's always nice to have another good citation (from
et al. and Gerstner).  I whacked a paragraph into the future work subsection
briefly discussing how this work could be put into NEAL.  It's the para
starting  "On the other hand,"

Oh, as the review is signed, I'd be interested to have a chat.  I can't actually see your signature.  Could you send me an email?