Pith. sign in

REVIEW 4 major objections 7 minor 39 references

Towards Applying Deep Learning to The Internet of Things: A Model and A Framework

T0 review · 4 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A six-objective weighted score can rank deep-learning optimization models for IoT devices, and a learned network can propose new ones when none match.

desk verdict An honest, well-scoped design-science proposal for a DL-optimization model-management framework, but it is an unvalidated artifact whose generative core needs a training corpus that is neither provided nor accounted for. read the letter →

arxiv 2501.06191 v1 pith:ZIFHBU2E submitted 2024-12-18 cs.NI

Samaa Elnagar , Kweku-Muata Osei-Bryson This is my paper · ORCID
classification cs.NI
keywords deeplearningInternetofThingsDLoptimizationmodelmanagementedgecomputingselectiondecisionsupportsystemtunneling
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that organizations embedding deep-learning networks in IoT devices lack criteria for choosing among competing optimization techniques such as pruning, quantization, and knowledge distillation. To fill that gap, it proposes a six-class modeling schema that captures the model itself, its optimization methods, cloud and end-device configurations, and performance measures. The schema feeds a management framework, DLOM2, that scores stored models by a weighted sum of performance, reliability, security, cost, latency, and complexity, and lets users pick the highest-scoring model. When no stored model matches, a DL Modeling Network is meant to learn from the repository and propose a new configuration. The paper positions this as an initial design-science artifact, not a deployed system.

What carries the argument

The load-bearing object is the six-class DL optimization modeling schema (Table 3), paired with the weighted-score formula in equation (1) and the DL Modeling Network. The schema defines what metadata about a model is recorded—network name, layers, hyperparameters, optimization methods, cloud configuration, end-device specs, and user ratings—which is what makes storage, retrieval, and comparison possible. Equation (1) collapses the six objectives into a single number for ranking. The DL Modeling Network is the only component that genuinely creates new models, and it is assumed to do so by learning patterns from a large repository of successful optimization models.

What would settle it

Try to assemble a corpus of successful DL optimization models with their device specs and outcomes; if no such corpus can be collected, or if a trained network produces no configuration that beats the best stored model on the same device, the framework's generative claim fails and only the scoring rule remains testable.

Watch

Extended reading notes

Core claim

The paper's central claim is that the DL optimization problem for IoT can be structured as a queryable, reusable decision process rather than an ad-hoc trial-and-error exercise. The proposed six-class schema—Model, Cloud Configuration, End-device Specifications, Main DLN, Optimization, and Performance—defines what metadata a stored optimization model must carry, enabling comparison and retrieval. Selection is driven by equation (1), a weighted sum of six objectives with weights derived from pairwise preference elicitation. The DLOM2 framework wraps this schema with a GUI, a cloud repository, a decision support system, and a DL Modeling Network; the last component is trained on previously stored models to infer new combinations of optimization techniques when the repository contains no match. The paper evaluates the framework only through a step-by-step illustrative example of a medical company, making clear that this is an initial design iteration.

Load-bearing premise

The framework's ability to create new models rests on training a DL Modeling Network on 'a huge number of successful optimization models,' but no such corpus is provided, sourced, or shown to exist; without it, DLOM2 is a query-and-rank engine over a database that is empty at present.

Editorial extensions

If this is right

  • If the schema is adopted, organizations can share and compare DL optimization models in a common format, lowering the cost of reusing prior work across projects.
  • The weighted-score rule gives IT managers a transparent, reproducible way to trade off accuracy, latency, cost, security, reliability, and complexity instead of relying on undocumented judgment.
  • When no stored model matches a request, the DL Modeling Network could synthesize a new configuration, making the repository grow with each query and user decision.
  • The paper's illustrative example shows a plausible workflow: capture requirements, query the repository, elicit preference weights, rank models, and optionally ask for a new model.
  • Because the design-science framing treats the artifact as an initial iteration, later work can revise the schema and scoring rule without discarding the overall management approach.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the DL Modeling Network proves infeasible to train for lack of data, the framework still operates as a weighted query engine, so its practical value depends on assembling a corpus before its generative promise can be tested.
  • The six-class schema could generalize beyond IoT to other model-selection settings, such as federated learning or model cards for transparency, because the same metadata pattern applies wherever models are chosen by context.
  • A concrete testable extension is to seed the repository with published compression results and check whether the weighted-score ranking matches expert choices in a user study.
  • Another testable extension is to measure how sensitive the top-ranked model is to the six weights, since pairwise preference weights can be noisy; the paper does not address that robustness question.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper addresses the problem of selecting and reusing Deep Learning (DL) optimization models for Internet of Things (IoT) applications. It proposes a six-class schema for describing DL optimization models (Model, Cloud Configuration, End Device Specifications, Main DLN, Optimization, and Performance classes) and an accompanying framework, DLOM2, composed of a cloud repository, a GUI, a decision support system (DSS), and a DL Modeling Network (DLN). The DSS ranks candidate models using a weighted additive score over six objectives (performance, reliability, security, cost, latency, complexity) as given in Equation (1). When no stored model matches the user criteria, the DLN is supposed to generate a new model configuration. The paper provides an illustrative example of the first three steps of the framework but includes no implementation, no data, and no evaluation.

Significance. If the proposed design were implemented and validated, the schema in Table 3 and the DSS workflow could provide a useful starting point for practitioners who need to choose among DL optimization techniques for edge and end devices. The paper usefully organizes known optimization methods around six explicit objectives and makes the selection procedure transparent and testable. Equation (1) is a standard multi-criteria aggregation, so there is no circularity in the ranking step itself. However, the paper's central claim, that the framework 'maximizes performance without sacrificing quality,' is not supported by any empirical evidence, and the generative DLN component depends on a training corpus that is never sourced. The main value of the paper at this stage is conceptual: the taxonomy and the decision workflow. The missing validation and the unspecified DLN training data are the key obstacles to accepting the stronger claims.

major comments (4)
  1. [§5.3 (DL Modeling Network) and §6 (Illustrative Example)] The DL Modeling Network is the only component that creates genuinely new optimization models, but the paper gives no source for the 'huge number of successful optimization models' it requires for training. Section 5.3 states the requirement without providing a corpus, citing an existing repository, or outlining an acquisition procedure. Because the DSS workflow queries the repository first and only invokes the DLN when no match is found, the entire generative claim in the abstract ('maximizes performance without sacrificing quality') depends on this missing training resource. The illustrative example in Section 6 never exercises the DLN, so it offers no evidence that such a network can be trained from the proposed schema. The authors should either supply a concrete data-source and feasibility argument, or explicitly restrict the paper's claims to retrieval and ranking of existing models.
  2. [§5.3 (DSS workflow)] The DSS saves 'a copy of the new model configurations in the repository' immediately after the DLN proposes a new model, without any measured performance or validation step. This would insert unvalidated predictions into the same repository that is supposed to contain 'successful optimization models' used as training input for the DLN. The feedback loop therefore degrades the training signal as soon as the system begins generating models. The paper needs an explicit provenance and validation mechanism, such as storing only models with measured results or tagging predicted versus validated models, before this component can support the framework's claims.
  3. [§6 and §7] The only evaluation is an illustrative example that walks through the GUI steps and applies Equation (1) to three retrieved models. It does not compare the framework against existing model management systems, measure any quality metric, test the DLN, or use a real repository. Section 7 itself concedes that the framework is 'an initial abstract design iteration.' The abstract's strong claim that the framework 'would help organizations choose the optimal DL optimization model that maximizes performance without sacrificing quality' is therefore not supported by the evidence presented. The claims need to be brought in line with the design-science stage, or a prototype and evaluation need to be added.
  4. [Equation (1)] Equation (1) is a standard weighted additive score: it ranks models according to the user's preference weights over six objectives. Consequently, the 'optimal' model is optimal only relative to the weights elicited from a particular user. The paper's wording that the framework selects an optimal model that 'maximizes performance without sacrificing quality' extends beyond what Equation (1) can deliver, since a weighted sum explicitly encodes trade-offs among performance, cost, latency, and so on. The claim should be scoped to 'the model ranked best under the user's stated preferences,' and the relationship between the weights and the performance/quality trade-off should be stated explicitly.
minor comments (7)
  1. [§5.3] The DLN is described as an 'unsupervised DL approach' but is also said to be 'trained with different optimization models previously stored in the repository'; the learning paradigm and the output of the network (a new configuration versus a ranking) should be clarified.
  2. [Equation (1)] The subscripts in Equation (1) are garbled in the submitted text (e.g., w89:, w;<=, w><?), and it is not stated whether the weights are normalized or how the reverse-scaled objectives (cost, latency, complexity) are combined with the forward-scaled ones; the notation should be fixed and the scale convention stated.
  3. [§6] Query 1 in Section 6 is not valid SPARQL as printed: the WHERE clause is not closed and the FILTER expression is incomplete; a correct query should be provided.
  4. [Table 2] Table 2 lists 'Design Evaluation' with the future-tense phrase 'will be evaluated using illustrative example,' which conflicts with the present-tense claims elsewhere in the paper and with the design-science methodology cited in Section 3; the research stage should be stated consistently.
  5. [§5.2] Section 5.2 describes four repositories but does not specify which of the six schema classes from Table 3 are stored in each; a mapping would help readers understand the storage design.
  6. [Throughout] There are numerous typographical and formatting issues, including 'a Deep Learning Networks (DLNs)' in Section 1, 'the research add s' in Section 1, 'thar' in Section 7, and inconsistent titles for the DLOM2 acronym; a careful copyedit is needed.
  7. [§5.3] The DSS paragraph is placed under the heading 'The DL Modeling Network' in Section 5.3; the subsection structure should separate the DSS description from the DLN description.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation; the framework is an under-specified design proposal, not a fitted prediction.

full rationale

The paper does not derive a quantitative result from fitted inputs. The closest thing to a selection rule is Equation (1), a weighted sum over the same six objectives (performance, reliability, security, cost, latency, complexity) that the schema defines as the criteria for optimality; this is a stated multi-criteria decision rule, not a hidden reduction, since the weights come from user preference elicitation rather than being fitted to the outcome being predicted. The DL Modeling Network (Section 5.3) is the only generative component, and the paper concedes it 'needs a huge number of successful optimization models to train such DL network'; no such corpus is provided or sourced. That is an unimplemented and unsupported design element, not a circular derivation, because no fitted model is claimed and no prediction is compared against training labels. The framework's modeling schema is said to be based on the DM3 ontology [13], a self-citation involving co-author Osei-Bryson, and [5] and [33] are also self-citations; however, none of these citations is used to establish the target claim that DLOM2 helps organizations choose optimal DL optimization models. Section 7 itself labels the work an 'initial abstract design iteration' that 'needs further work to be completed,' reinforcing that the framework is incomplete rather than self-validating. No equation or definition reduces to its own input, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 1 free parameters · 4 assumptions · 1 invented entities

The framework's operation rests on several unstated or weakly supported assumptions: the adequacy of DM3 for the schema, the validity of subjective user ratings, the existence of a training corpus for the DLN, and the sufficiency of a simple additive weighting rule. None of these are tested.

free parameters (1)
  • Preference weights w_perf, w_rel, w_sec, w_cst, w_lat, w_cmp
    Equation (1) determines model ranking by these weights. They are chosen by the user during Preference Elicitation (Section 6) and are not grounded in data or derived from any measurement.
assumptions (4)
  • domain assumption The DM3 ontology provides a sound basis for the DL optimization schema.
    Section 4.1 builds the schema 'based on the established model of DM3 ontology' [13] with no evaluation of fit or completeness.
  • domain assumption Pairwise comparison preference elicitation yields valid objective weights.
    Used in Section 6 without validation, relying on reference [36]; the paper does not test whether the weights reflect actual user priorities.
  • domain assumption User ratings on a 1 to 5 scale for six objectives represent trustworthy model quality.
    The Model Class 'Rating' attribute is the core selection input in the schema (Table 3); there is no check of inter-rater reliability, bias, or consistency across users.
  • ad hoc to paper A sufficient repository of successful optimization models exists or will exist to train the DL Modeling Network.
    Section 5.3 states the DLN 'needs a huge number of successful optimization models to train such DL network'; this corpus is neither provided nor sourced.
invented entities (1)
  • DL Modeling Network (DLN)
    purpose: Unsupervised model that analyzes stored optimization models to recommend or create new DLN configurations for IoT
    Proposed in Section 5.3 as the 'prediction platform' component; it is not implemented, has no training data, and offers no falsifiable handle outside the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Applying Deep Learning to The Internet of Things: A Model and A Framework." pith.science (2026). https://pith.science/paper/ZIFHBU2E

@misc{pith2026250106191,
  author       = {Pith},
  title        = {Pith review of: Towards Applying Deep Learning to The Internet of Things: A Model and A Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZIFHBU2E}},
  note         = {Machine review of arXiv:2501.06191}
}
read the original abstract

Deep Learning (DL) modeling has been a recent topic of interest. With the accelerating need to embed Deep Learning Networks (DLNs) to the Internet of Things (IoT) applications, many DL optimization techniques were developed to enable applying DL to IoTs. However, despite the plethora of DL optimization techniques, there is always a trade-off between accuracy, latency, and cost. Moreover, there are no specific criteria for selecting the best optimization model for a specific scenario. Therefore, this research aims at providing a DL optimization model that eases the selection and re-using DLNs on IoTs. In addition, the research presents an initial design for a DL optimization model management framework. This framework would help organizations choose the optimal DL optimization model that maximizes performance without sacrificing quality. The research would add to the IS design science knowledge as well as the industry by providing insights to many IT managers to apply DLNs to IoTs such as machines and robots.

Figures

Figures reproduced from arXiv: 2501.06191 by the authors.

Figure 4
Figure 4. the ontological representation DL Optimization Model Schema [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. The DLOM2 Architecture [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 36 canonical work pages

  1. [1]

    To build an DL optimization model that provides an end-to-end modeling for ap- plying DLN to IoTs according to specific contextual settings

    work page

  2. [2]

    computing devices often with sensor capability to collect, share, and transfer data using the Internet

    To provide the foundational bricks for building a De ep Learning Optimization Model Management Framework (DLOM)2 that maximizes value gained from ap- plying the presented model. The research add s to IS body of knowledge by creating an optimization model that connects two emergent yet important paradigms: DL and IoT. In addition, the paper sheds light on ...

    work page

  3. [3]

    Deep Learning in Real Time – Inference Acceleration and Continuous Training

    Synced. Deep Learning in Real Time – Inference Acceleration and Continuous Training . 2017

    work page 2017

  4. [4]

    IEEE Journal of Biomedical and Health Informatics, 2019

    Li, K., et al., Convolutional recurrent neural networks for glucose p rediction. IEEE Journal of Biomedical and Health Informatics, 2019

    work page 2019

  5. [5]

    Wu, J., et al., PocketFlow: An Automated Framework for Compressing and Ac celerating Deep Neural Networks. 2018

    work page 2018

  6. [6]

    IEEE Internet of Things Journal, 2016

    Shi, W., et al., Edge computing: Vision and challenges. IEEE Internet of Things Journal, 2016. 3(5): p. 637-646

    work page 2016

  7. [7]

    Behaviour & Information Technology, 2013

    Boos, D., et al., Controllable accountabilities: the internet of thing s and its challenges for organisations. Behaviour & Information Technology, 2013. 32(5): p. 449-467

    work page 2013

  8. [8]

    Elnagar, S. and M. Thomas, Federated Deep Learning: A Conceptual Model and Applied Framework for Industry 4.0 in Americas Conference on Information Systems 2020 . 2020, AIS

    work page 2020

Show all 39 references
  1. [9]

    IEEE Transactions on Industrial Informatics, 2019

    Lyu, L., et al., Fog-embedded Deep Learning for the Internet of Things. IEEE Transactions on Industrial Informatics, 2019

    2019

  2. [10]

    Automated Management of Deep Learning Experiments

    Gharibi, G., et al. Automated Management of Deep Learning Experiments . in Proceedings of the 3rd International Workshop on Data Management for End -to-End Machine Learning. 2019. ACM

    2019

  3. [11]

    Towards unified data and lifecycle management for de ep learning

    Miao, H., et al. Towards unified data and lifecycle management for de ep learning . in 2017 IEEE 33rd International Conference on Data Engineering (ICDE) . 2017. IEEE

    2017

  4. [12]

    Ota, and M

    Li, H., K. Ota, and M. Dong, Learning IoT in edge: Deep learning for the Internet of Things with edge computing. IEEE Network, 2018. 32(1): p. 96-101. 14

    2018

  5. [13]

    Procedia computer science, 2017

    O’Connor, Y., et al., Privacy by design: informed consent and internet of things for smart health. Procedia computer science, 2017. 113: p. 653-658

    2017

  6. [14]

    Lunardi, W.T., Decision support IoT framework: device discovery and stream analytics. 2016

    2016

  7. [15]

    Han, and K

    Ko, S.-W., K. Han, and K. Huang, Wireless networks for mobile edge computing: Spatial modeling and latency analysis. IEEE Transactions on Wireless Communications, 2018. 17(8): p. 5225-5240

    2018

  8. [16]

    Thomas, and K

    Li, Y., M.A. Thomas, and K. -M. Osei-Bryson, Ontology-based data mining model management for self-service knowledge discovery. Information Systems Frontiers, 2017. 19(4): p. 925-943

    2017

  9. [17]

    Software: Practice and Experience, 2017

    Gupta, H., et al., iFogSim: A toolkit for modeling and simulation of resource manage ment techniques in the Internet of Things, Edge and Fog computin g environments. Software: Practice and Experience, 2017. 47(9): p. 1275-1296

    2017

  10. [18]

    arXiv preprint arXiv:1902.04413, 2019

    Kunkel, R., et al., TensorSCONE: A secure TensorFlow framework using Intel SGX. arXiv preprint arXiv:1902.04413, 2019

    1902 arXiv

  11. [19]

    M odel DB: a system for machine learning model manag ement

    Vartak, M., et al. M odel DB: a system for machine learning model manag ement. in Proceedings of the Workshop on Human -In-the-Loop Data Analytics. 2016. ACM

    2016

  12. [20]

    Automatically tracking metadata and provenance of ma chine learning experiments

    Schelter, S., et al. Automatically tracking metadata and provenance of ma chine learning experiments. in Machine Learning Systems workshop at NIPS . 2017

    2017

  13. [21]

    Deep fried convnets

    Yang, Z., et al. Deep fried convnets . in Proceedings of the IEEE International Conference on Computer Vision. 2015

    2015

  14. [22]

    Cryptonets: Applying neural networks to encrypted data with high throughput and accuracy

    Gilad-Bachrach, R., et al. Cryptonets: Applying neural networks to encrypted data with high throughput and accuracy. in International Conference on Machine Learning . 2016

    2016

  15. [23]

    apparea” is the business focus and “cost

    is another compression technique that involves training a quantized neural network (student model) with the help of a full -precision pre-trained network (teacher model). The compressed student model can take the benefit of transferring knowledge from the teacher model. Quantiz...

  16. [24]

    Hanson, S.J. and L.Y. Pratt. Comparing biases for minimal network construction with back - propagation. in Advances in neural information processing systems . 1989

    1989

  17. [25]

    Alsing, O., Mobile Object Detection using TensorFlow Lite and Tr ansfer Learning. 2018

    2018

  18. [26]

    Vinyals, and J

    Hinton, G., O. Vinyals, and J. Dean, Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015

    2015 arXiv

  19. [27]

    Mao, and W.J

    Han, S., H. Mao, and W.J. Dally, Deep compression: Compressing deep neural networks wit h pruning, trained quantization and huffman coding. arXiv preprint arXiv:1510.00149, 2015

    2015 arXiv

  20. [28]

    arXiv preprint arXiv:1812.07872, 2018

    Goncharenko, A., et al., Fast adjustable threshold for uniform neural network quantization. arXiv preprint arXiv:1812.07872, 2018

    2018 arXiv

  21. [29]

    Quantization and training o f neural networks for efficient integer -arithmetic-only inference

    Jacob, B., et al. Quantization and training o f neural networks for efficient integer -arithmetic-only inference. in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018

    2018

  22. [30]

    Wang, and W

    Zhang, X., Y. Wang, and W. Shi. pcamp: Performance comparison of machine learning packages on the edges. in {USENIX} Workshop on Hot Topics in Edge Computing (HotEdge 1 8). 2018

    2018

  23. [31]

    arXiv preprint arXiv:1905.00571, 2019

    Niu, W., et al., 26ms Inference Time for ResNet-50: Towards Real-Time Execution of all DNNs on Smartphone. arXiv preprint arXiv:1905.00571, 2019

    1905 arXiv

  24. [32]

    arXiv preprint arXiv:1811.01395, 2018

    Bhunia, A.K., et al., Query-based Logo Segmentation. arXiv preprint arXiv:1811.01395, 2018

    2018 arXiv

  25. [33]

    Accelerating Deep Learning inference on mobile system s

    Frajberg, D., et al. Accelerating Deep Learning inference on mobile system s. in International Conference on AI and Mobile Services . 2019. Springer

    2019

  26. [34]

    Gregor, S. and A.R. Hevner, Positioning and presenting design science research for maximum impact. MIS quarterly, 2013: p. 337 -355

    2013

  27. [35]

    Chang, S. -k. Information exchange theory and man -machine interaction . in 1982 21st IEEE Conference on Decision and Control . 1982. IEEE

    1982

  28. [36]

    Yoon, and M

    Elnagar, S., V. Yoon, and M. Thomas. An Automatic Ontology Generation Framework with An Organizational Perspective . in Proceedings of the 53rd Hawaii International Confere nce on System Sciences. 2020

    2020

  29. [37]

    MIS quarterly, 2004

    Hevner, A., et al., Design science research in information s ystems. MIS quarterly, 2004. 28(1): p. 75-105

    2004

  30. [38]

    What is measurement stability

    Minitab. What is measurement stability . 2018; Available from: https://support.minitab.com/en - us/minitab/18/help-and-how-to/quality-and-process-improvement/measurement -system- analysis/supporting-topics/other-gage-studies-and-measures/what-is-measurement-stability/

    2018

  31. [39]

    Chen, L. and P. Pu, Survey of preference elicitation methods . 2004

    2004

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.