Write a comment

Does the introduction explain the objective of the research presented in the preprint?

Yes

The introduction clearly states that the paper aims to present the historical and evolutionary development of deep learning by linking major breakthroughs to the limitations they addressed and to their influence on subsequent architectures. The authors also clarify that the paper is not intended as an exhaustive survey, but as a selective discussion of seminal methods that established new research directions.

Are the methods well-suited for this research?

Somewhat inappropriate

A narrative historical review is suitable in principle, but the preprint does not provide a reproducible review methodology. It does not explain how sources were searched, screened, or selected, or define the criteria used to classify papers as “seminal,” “representative,” or “highly influential.” The authors explicitly adopt a selective rather than exhaustive approach, but the basis for that selection is not reported. The manuscript also proceeds directly from the introduction to architecture-specific sections without a dedicated methods section. This creates a substantial risk of subjective selection and incomplete coverage. Nevertheless, the chronological organization and discussion of landmark studies provide some reasonable basis for addressing the broad descriptive aim, so “Highly inappropriate” would be too severe.

Are the conclusions supported by the data?

Somewhat unsupported

The manuscript is a narrative historical review rather than an empirical study, so its conclusions depend on the selection and interpretation of previously published literature. The authors conclude that they have presented the major turning points in the evolution of deep learning and suggest that the review will help the research community rapidly understand the field’s development. These conclusions are broadly consistent with the topics covered in the manuscript, including neural networks, CNNs, RNNs, transformers, GANs, diffusion models, and application-specific architectures. However, the strength of these conclusions cannot be fully assessed because the manuscript does not describe how the reviewed literature was identified or selected. No databases, search terms, date limits, inclusion or exclusion criteria, screening procedure, or quality-assessment process are reported. The authors state that they focus on “seminal methods” and “representative applications,” but they do not define these categories or provide transparent criteria for determining which studies qualify. Consequently, it is not possible to establish whether the selected publications provide a complete, balanced, or reproducible representation of the historical development of deep learning. The conclusion also claims that the review will help the research community identify the major turning points in the field. This is plausible, but it is presented as an expectation rather than a conclusion demonstrated by the review. Overall, the conclusions are generally compatible with the manuscript’s descriptive content, but claims concerning comprehensiveness, representativeness, and usefulness are not sufficiently supported by a documented review methodology.

Are the data presentations, including visualizations, well-suited to represent the data?

Somewhat inappropriate or unclear

The visualizations are relevant to the manuscript’s historical-review format, but their execution substantially limits readability and interpretability. Figure 1 presents a useful chronological overview, but it contains many model names, dates, and architectural categories in a relatively small space. The labels are very small, several entries are tightly clustered around 2014–2023, and the light-colored lines provide limited contrast. This makes the timeline difficult to read without considerable magnification. The caption identifies the architectural categories but does not explain the criteria used to select the displayed models or why particular milestones were included while others were omitted. Figure 2 is even more problematic. It combines numerous neural-network, CNN, RNN, transformer, GAN, and diffusion architectures into one highly compressed composite figure. Individual labels, mathematical symbols, arrows, and legends are too small for comfortable reading at normal page size. The caption merely describes it as a visual depiction of architectural frameworks and does not provide sufficient guidance for interpreting the individual diagrams or their relationships. Table 1 contains useful bibliographic and architectural information, but it is extremely dense and uses numerous abbreviations. More importantly, the single “Performance” column combines fundamentally different metrics—such as accuracy, mAP, mIoU, WER, perplexity, MSE, and FVD—measured on different datasets and tasks. These values are therefore not directly comparable, yet the presentation does not clearly warn readers against cross-row comparisons. Some parameter and performance fields are also absent or inconsistently reported. The presentations are not entirely unhelpful: the timeline conveys the general chronology, the diagrams illustrate the diversity of architectures, and the table compiles information on representative applications. However, they do not meet a high standard of clarity or accessibility. They are overcrowded, require substantial magnification, do not explain the criteria used to select the material, and provide insufficient context for comparing the reported results.

How clearly do the authors discuss, explain, and interpret their findings and potential next steps for the research?

Neither clearly nor unclearly

The manuscript explains individual developments chronologically, but the overall interpretation is limited. The conclusion is very brief, does not discuss limitations or broader lessons, and does not provide concrete future research directions despite promising them in the introduction.

Is the preprint likely to advance academic knowledge?

Not likely

The preprint may be useful as an introductory historical overview, but it primarily summarizes established architectures and previously published milestones. It does not present new empirical findings, a novel theoretical framework, a validated taxonomy, or a reproducible systematic-review method. The authors explicitly describe the paper as a selective discussion of “seminal methods” and representative applications. Its stated contribution is mainly to help readers quickly review major turning points in deep learning, rather than to generate or confirm new academic knowledge.

Would it benefit from language editing?

Yes

The manuscript would benefit from language editing. It contains grammatical errors, awkward phrasing, inconsistent terminology, and typographical mistakes, such as “Backpropogation,” “Parallely stacks,” and the unclear phrase “highlighted the importance of the proposition.” These issues do not make the paper unintelligible, but they reduce precision, professionalism, and reading fluency.

Would you recommend this preprint to others?

No, it’s of low quality or is majorly flawed

The manuscript provides a broad descriptive overview, but its scholarly value is limited by the absence of a transparent review methodology, undefined criteria for selecting “seminal” studies, weak critical synthesis, overcrowded visualizations, and an underdeveloped conclusion. These are substantial methodological and presentation weaknesses rather than minor issues, so the preprint cannot currently be recommended to others as a reliable academic review.

Is it ready for attention from an editor, publisher or broader audience?

No, it needs a major revision

The manuscript requires major revision before it is ready for editorial, publishing, or broader-audience attention. Key issues include the absence of a transparent and reproducible review methodology, unclear criteria for selecting influential studies, limited critical synthesis, overcrowded and inconsistently presented visual material, weak discussion of limitations and future directions, and numerous language problems. The conclusion is also too brief to support the broad claims made by the authors.

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.