PCSY-D-25-00006

Predictive Coding algorithms induce brain-like responses in Artificial Neural Networks

PLOS Complex Systems

Dear Dr. Gütlin,

Thank you for submitting your manuscript to PLOS Complex Systems. After careful consideration, we feel that it has merit but does not fully meet PLOS Complex Systems's publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript within 60 days May 31 2025 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at complexsystems@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pcsy/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

* A rebuttal letter that responds to each point raised by the editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. This file does not need to include responses to any formatting updates and technical items listed in the 'Journal Requirements' section below.

* A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

* An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, competing interests statement, or data availability statement, please make these updates within the submission form at the time of resubmission. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

We look forward to receiving your revised manuscript.

Kind regards,

Juan Gonzalo Barajas-Ramirez

Academic Editor

PLOS Complex Systems

Juan Gonzalo Barajas-Ramirez

Academic Editor

PLOS Complex Systems

Hocine Cherifi

Editor-in-Chief

PLOS Complex Systems

Journal Requirements:

1. We ask that a manuscript source file is provided at Revision. Please upload your manuscript file as a .doc, .docx, .rtf or .tex.

Additional Editor Comments (if provided):

AE: The reviewers find that the manuscript addresses an interesting topic, yet are not completely satisfied by the results given. The originality and accuracy of the model approach for to simulate brain activity needs further details and to be make more tractable. The authors are encourage to prepare a revised version of the manuscript with a detail point by point answer to the comments and corrections required by the reviewers.

Reviewer 1:

The paper presents an interesting comparison of predictive coding, contrastive learning and backprop to identify which learning objective is more aligned with biological-plausible learning. The results are rigorously analyzed using statistical measures, I have the following concerns regarding the work.

Major concerns:

1. Originality -- the learning objectives compared here are already existing and it has been proven the PC-based learning algorithms are better at biologically plausible learning in DNNs. I am not quite sure what original contribution is being made here specially with an easily learnable dataset of MNIST (even with the augmentations). In the choice of the model, you discuss that you use a SimpleRNN model which is "the most minimalistic and assumption-free type of network capable

of expressing prior predictions" as oppose to PredNet. Is the originality of the work related to the application of PC-based learning objectives to such simple RNN models?

2. Accuracy of claims -- in the introduction, when formulating the problem, you iterate the phrase "biological neural networks". However, the cited works [3-8] talks about the biological-plausible neural network. Usually the term "biological neural networks" are used to refer to computational networks where the dynamics are defined using differential equations. Further, you claim that supervision is not used in the brain learning mechanisms, however, the brain create associations between different attributes of the same objective using the supervision. Further, there are spike neural networks that are inspired from brain dynamic equations which mimics how spatio-temporal learning happens in the brain treating the neurons of the NN as biological neurons. You may need to rephrase some of these claims to align with the current state of the art.

Minor concerns:

1. Figures - It is better to increase the font sizes, specifically with the legend entries in Figure 3/4.

While the claims based on your simulations are great, I am not clear about the original contribution of this work. After you fix the content where precise and clear indication is given about the contributions of this paper as opposed to the already established knowledge, I think the paper is in a good shape to be accepted.

Reviewer 2:

In this manuscript, the authors provide two complimentary ways to simulate brain activity via contrastive learning and predictive coding. They find the two methods have some realism compared to how real brains function. I, however, have a few critiques:

• The authors do not have code available. For reproducibility, and so I could assess the method, I would like to have access to the code within your Github link.

• MNIST can be separated even with clustering – I would like to see the robustness of results with other datasets, e.g., FashionMNIST, CIFAR-10, etc.

• It would be useful to assess the robustness of your results with different hyperparameters

• Note that because of grokking, you might find larger models exhibit better training behavior (including semantic encoding, etc.) than smaller models, so it would be good to compare with models of larger size to assess whether the model simulates brains that are orders of magnitude more complex.

• The authors use very simple models. What about transformers? Other methods? It seems there are a number of others ways one could model the brain.

• I would also like to see more details about the models. How much training occurred? Was there early stopping?

• As alluded to in 3.2, Comparisons seem too simplistic, and do not necessarily represent a model of brains. The current LLMs could do better, and are based on often very different training methods. Does that mean they are a better representation of the brain? Given the very large differences between the models and real brains, it is therefore hard to assess if these results really do offer insights into the emergent behavior of brains.

• The authors mention BrainScore - why don't they use it? – namely contrastive/predictive language models can be used on BrainScore language models. I am assuming here that a model to understand language should be similar to one that can understand image sequences, as its all fundamentally prediction, so why use a different model like image prediction?

• Finally, why were the image movement benchmarks created adequate? There are a number of ways to simulate movement including real videos. The current method of shifting numbers around does not feel sufficiently ecologically valid.

[Note: HTML markup is below. Please do not edit.]

Reviewers' Comments:

Reviewer's Responses to Questions

Comments to the Author

1. Does this manuscript meet PLOS Complex Systems’s publication criteria?>

Reviewer #1: Partly

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?-->?>

Reviewer #1: Yes

Reviewer #2: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available (please refer to the Data Availability Statement at the start of the manuscript PDF file)??>

The PLOS Data policy

Reviewer #1: Yes

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English??>

Reviewer #1: Yes

Reviewer #2: Yes

**********

Reviewer #1: The paper presents an interesting comparison of predictive coding, contrastive learning and backprop to identify which learning objective is more aligned with biological-plausible learning. The results are rigorously analyzed using statistical measures, I have the following concerns regarding the work.

Major concerns:

1. Originality -- the learning objectives compared here are already existing and it has been proven the PC-based learning algorithms are better at biologically plausible learning in DNNs. I am not quite sure what original contribution is being made here specially with an easily learnable dataset of MNIST (even with the augmentations). In the choice of the model, you discuss that you use a SimpleRNN model which is "the most minimalistic and assumption-free type of network capable

of expressing prior predictions" as oppose to PredNet. Is the originality of the work related to the application of PC-based learning objectives to such simple RNN models?

2. Accuracy of claims -- in the introduction, when formulating the problem, you iterate the phrase "biological neural networks". However, the cited works [3-8] talks about the biological-plausible neural network. Usually the term "biological neural networks" are used to refer to computational networks where the dynamics are defined using differential equations. Further, you claim that supervision is not used in the brain learning mechanisms, however, the brain create associations between different attributes of the same objective using the supervision. Further, there are spike neural networks that are inspired from brain dynamic equations which mimics how spatio-temporal learning happens in the brain treating the neurons of the NN as biological neurons. You may need to rephrase some of these claims to align with the current state of the art.

Minor concerns:

1. Figures - It is better to increase the font sizes, specifically with the legend entries in Figure 3/4.

While the claims based on your simulations are great, I am not clear about the original contribution of this work. After you fix the content where precise and clear indication is given about the contributions of this paper as opposed to the already established knowledge, I think the paper is in a good shape to be accepted.

Reviewer #2: In this manuscript, the authors provide two complimentary ways to simulate brain activity via contrastive learning and predictive coding. They find the two methods have some realism compared to how real brains function. I, however, have a few critiques:

• The authors do not have code available. For reproducibility, and so I could assess the method, I would like to have access to the code within your Github link.

• MNIST can be separated even with clustering – I would like to see the robustness of results with other datasets, e.g., FashionMNIST, CIFAR-10, etc.

• It would be useful to assess the robustness of your results with different hyperparameters

• Note that because of grokking, you might find larger models exhibit better training behavior (including semantic encoding, etc.) than smaller models, so it would be good to compare with models of larger size to assess whether the model simulates brains that are orders of magnitude more complex.

• The authors use very simple models. What about transformers? Other methods? It seems there are a number of others ways one could model the brain.

• I would also like to see more details about the models. How much training occurred? Was there early stopping?

• As alluded to in 3.2, Comparisons seem too simplistic, and do not necessarily represent a model of brains. The current LLMs could do better, and are based on often very different training methods. Does that mean they are a better representation of the brain? Given the very large differences between the models and real brains, it is therefore hard to assess if these results really do offer insights into the emergent behavior of brains.

• The authors mention BrainScore - why don't they use it? – namely contrastive/predictive language models can be used on BrainScore language models. I am assuming here that a model to understand language should be similar to one that can understand image sequences, as its all fundamentally prediction, so why use a different model like image prediction?

• Finally, why were the image movement benchmarks created adequate? There are a number of ways to simulate movement including real videos. The current method of shifting numbers around does not feel sufficiently ecologically valid.

**********

what does this mean? ). If published, this will include your full peer review and any attached files.

Do you want your identity to be public for this peer review? If you choose “no”, your identity will remain anonymous but your review may still be made public.

For information about this choice, including consent withdrawal, please see our Privacy Policy

Reviewer #1: No

Reviewer #2: No

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

Figure resubmission:

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step. If there are other versions of figure files still present in your submission file inventory at resubmission, please replace them with the PACE-processed versions.

Reproducibility:

To enhance the reproducibility of your results, we recommend that authors of applicable studies deposit laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Predictive Coding algorithms induce brain-like responses in Artificial Neural Networks

PCSY-D-25-00006R1

Dear Dr. Gütlin,

We're pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you'll receive an e-mail detailing the required amendments. When these have been addressed, you'll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at https://www.editorialmanager.com/pcsy/ click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. For questions related to billing, please contact billing support at https://plos.my.site.com/s/.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they'll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact complexsystems@plos.org.

Kind regards,

Juan Gonzalo Barajas-Ramirez

Academic Editor

PLOS Complex Systems

Additional Editor Comments (optional):

Reviewer #1:

Reviewer #3:

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

Reviewer #1: All comments have been addressed

Reviewer #3: All comments have been addressed

--------------------

publication criteria?>

Reviewer #1: Yes

Reviewer #3: Yes

--------------------

3. Has the statistical analysis been performed appropriately and rigorously?-->?>

Reviewer #1: Yes

Reviewer #3: Yes

--------------------

4. Have the authors made all data underlying the findings in their manuscript fully available (please refer to the Data Availability Statement at the start of the manuscript PDF file)??>

The PLOS Data policy

Reviewer #1: Yes

Reviewer #3: Yes

--------------------

5. Is the manuscript presented in an intelligible fashion and written in standard English?<br/><br/>PLOS Complex Systems does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.-->?>

Reviewer #1: Yes

Reviewer #3: Yes

--------------------

Reviewer #1: The authors have addressed the previous comments related to the novelty/objectivity of the work, the accuracy regarding some terminology and phrases. With the changes added to abstract, introduction, and Methodology (with the help of appendices), the objective of the research is clear and understood as methodologically analyzing how predictive coding mimics the actual mechanisms of the brain during learning as oppose to the supervised backprop. Therefore, I would like to recommend the work to be accepted in its revised form.

Reviewer #3: The authors prepared a revised version where the previous comments and recommendations are addressed. Clarifications of the contribution and the sense in which Predicative coding algorithms are implemented as well as additional appendices that clarify the contribution are added to the revised version. One minor comment is the authors of reference 14 are only capital letters perhaps this needs to be corrected.

--------------------

what does this mean? ). If published, this will include your full peer review and any attached files.

Do you want your identity to be public for this peer review? If you choose “no”, your identity will remain anonymous but your review may still be made public.

For information about this choice, including consent withdrawal, please see our Privacy Policy

Reviewer #1: No

Reviewer #3: No

--------------------