The DOI matches its Crossref record; 127 of the 129 references it cites were found in a registry or an index; 56 of the 158 sources its claims cite were checked for support.
A Comprehensive Survey on TinyML
Open accessAbstract
Recent spectacular progress in computational technologies has led to an unprecedented boom in the field of Artificial Intelligence (AI). AI is now used in a plethora of research areas and has demonstrated its capability to bring new approaches and solutions to various research problems. However, the extensive computation required to train AI algorithms comes with a cost. Driven by the need to reduce the energy consumption, the carbon footprint and the cost of computers running machine learning algorithms, TinyML is nowadays considered as a promising AI alternative focusing on technologies and applications for extremely low-profile devices. This paper presents the results of a literature survey of all TinyML applications and related research efforts. Our survey builds a taxonomy of TinyML techniques that have been used so far to bring new solutions to various domains, such as healthcare, smart farming, environment, and anomaly detection. Finally, this survey highlights the remaining challenges and points out possible future research directions. We anticipate that this survey will motivate further discussions on the various fields of applications of TinyML and the synergy of resource-constrained devices and edge intelligence.
Overview
CoChat Summary
TL;DR
This survey reviews TinyML applications and research, builds a taxonomy across environment, smart farming, healthcare, and anomaly detection, and highlights benefits and challenges. It argues TinyML enables low-latency, low-power, private edge ML on microcontrollers but faces memory, accuracy, privacy, heterogeneity, reliability, and concept-drift challenges.
Key findings
- TinyML benchmark comparison reports latency of 0.18–300 ms, power consumption of 25–300 mW, accuracy of 80–90%, and memory consumption of few KB; it has very high privacy and does not rely on network connectivity, versus cloud computing at 10–500 ms, 50–1000 W, 86–94% accuracy, and GBs–TBs memory.
- The survey’s taxonomy covers environment, smart farming, healthcare, and anomaly detection; in smart farming, the most deployed case is crop management with 8 papers, followed by smart greenhouse with 3 papers and smart irrigation with 2 papers.
- Reported application results include mosquito wing-beat classification at 88.3% accuracy, ECG reservoir computing at 95.4% accuracy with variance 0.001, strawberry greenhouse MLP at 97% validation and 96% test accuracy with 151 trainable parameters, road anomaly TEDA at 99% accuracy and 0.76 F1-score, thermal-image predictive maintenance at 94% accuracy and 92% F1-score, and washing-machine autoencoder at 92% accuracy and 90% precision.
- Challenges identified include limited memory, accuracy drop (e.g., from 80% to 78.4%, and a 1.3% drop), privacy, hardware/software heterogeneity, trustworthiness/reliability, and concept drift; on-device/online learning such as TinyOL is discussed as a response.
Methods
- The paper uses a Systematic Literature Review (SLR) procedure: defining research questions, retrieving the literature, evaluating the literature, extracting data, and presenting/discussing results.
- Literature was retrieved from IEEE Xplore Digital Library, Science Direct, Springer Link, and Google Scholar, using terms including TinyML, Tiny-ML, embedded, tiny machine learning, embedded machine learning, edge intelligence, and TinyDL.
- The survey is guided by three research questions: new AI publications using tiny devices, smart solutions mitigating existing AI limitations, and trending topics, remaining challenges, and future research directions; sectors were chosen based on publication volume: environment, smart farming, healthcare, and anomaly detection.
Limitations
- The paper states TinyML challenges including limited memory, accuracy drop, privacy, heterogeneity of hardware and software, product trustworthiness and reliability, and concept drift; it also notes the TinyML benchmark is still in early stages, with the latest version being 1.0.
- Evident from the text: as a survey, it does not conduct new experiments or generate primary data; its findings aggregate prior works, and some summary tables mark missing information with “-”, so reported metrics depend on the cited papers’ setups.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
26 Valid · 38 Not valid · 91 Not checked · 30 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“After deploying the data collect by sensors on trained ML models, [107] obtained an accuracy that ranges from 53.6% to 69.9%, in contrast, [106] achieved an accuracy of 69% to 73.08%”
§ A Comprehensive Survey on TinyML > D. EMOTION DETECTION
What the sources say
- Disputedcites [106] sourceExperimental — accuracy not yet measured
Table 5 in the Final considerations reports accuracy for classifying mosquitoes by wingbeat sound, not for emotion recognition. Its reported accuracies are 97.65% for the binary model, 78.12% for the multiclass model, and 94.56% for the ensemble—not 69% to 73.08%. (LLM, conf=0.90)
Checked by a reasoning model — may be wrong.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“The authors in [126] developed a multi-label TinyML model based on Multi-Layer Perceptron(MLP) architecture for micro-climate control of a strawberry agricultural greenhouse.”
§ A Comprehensive Survey on TinyML > C. SMART GREENHOUSE
What the sources say
- Disputedcites [126] sourceExperimental — accuracy not yet measured
Section 4.3.1 describes a microcontroller model for reading numerical water meters, not for controlling the micro-climate of a strawberry greenhouse. It says AutoKeras was used to find the network architecture; it does not identify the model as an MLP. (LLM, conf=0.88)
Checked by a reasoning model — may be wrong.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“This device helps farmers to monitor greenhouse gases coming from the soil.”
§ A Comprehensive Survey on TinyML > C. SMART GREENHOUSE
What the sources say
- Disputedcites [125] sourceExperimental — accuracy not yet measured
Sections 4.2 and 4.3.1 describe sensors used to assess plants’ moisture and nutrient needs and guide irrigation or fertigation. Section 4.3.1 specifies the monitored parameters as air temperature, soil moisture, soil pH, and soil electrical conductivity—not greenhouse gases emitted from soil. No page numbers are available in our copy. (LLM, conf=0.86)
Checked by a reasoning model — may be wrong.
plainstates a causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“These customized devices can attain performance in the one micro Joule per inference level, pushing the limits of machine learning to the ultra-low power end of TinyML processors.[48]”
§ A Comprehensive Survey on TinyML > C. HARDWARE CONSTRAINTS
What the sources say
- Partly supportedcites [48] Hosseininoorbin, 2021, Exploring edge TPU for network intrusion detection in IoT, arXiv:2103.16295 sourceExperimental — accuracy not yet measured
Section 4.2.1 (p. 8) describes the Edge TPU as a purpose-built ML accelerator with a low energy footprint, and Section 5.1.3 (p. 12) evaluates energy efficiency per inference. But the passages do not establish the claimed one-microjoule-per-inference level or characterize the devices as ultra-low-power TinyML processors; the paper’s 1 mJ figure is an energy budget used to define its efficiency measure, not energy per inference. (LLM, conf=0.82)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“The market has seen an emergence of the new 32bits generation of IoT-ready microcontrollers [47].”
§ A Comprehensive Survey on TinyML > C. HARDWARE CONSTRAINTS
What the sources say
- Could not verifycites [47] sourceExperimental — accuracy not yet measured
The Introduction mentions memory-limited IoT edge devices, including microcontrollers, but does not discuss a new 32-bit generation or a market trend. Searches for “32-bit generation of IoT-ready microcontrollers” and “microcontroller 32-bit IoT market” found no passage addressing that claim. (LLM, conf=0.75)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
44 more claims · 13 with their cited sources checked
Create a free account to see all 185 claims and to verify the 1 reference still open.
- Probably not a substantial claim. 104
- It may state two separate claims as one. 74
- The quote only partly states this claim. 1
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ A Comprehensive Survey on TinyML > I. INTRODUCTION
“For ease of use, the acronyms are summarized in Table. 5.”
§ A Comprehensive Survey on TinyML > Additional information
“Table: 9 TABLE 9 . Table of Acronyms.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ A Comprehensive Survey on TinyML > B. TINYML
“The values presented in the table are derived from the benchmark conducted by MLCommons [45], an open engineering consortium dedicated to advancing machine learning innovation and fostering the development of cutting-edge ML models and systems.”
§ A Comprehensive Survey on TinyML > References
“45. "Dawnbench." [Online]. Available: https://dawn.cs.stanford.edu/benchmark/CIFAR10/inference.html”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ A Comprehensive Survey on TinyML > C. INTERNET OF INTELLIGENT VEHICLES
“The results on real experiments (Asphalt pavement) show an accuracy of 99% and an f1-score of 0.76.”
§ A Comprehensive Survey on TinyML > VIII. CONCLUSION
“| [141] | Detecting road anomalies | - | TEDA | Arduino BLE Sense Nano 33 | TensorFlow | 98 | 54 mW | 11 ms | - |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ A Comprehensive Survey on TinyML > A. CROP MANAGEMENT
“The authors in [96] built a prototype to detect and count the number of strawberries in the image.”
§ A Comprehensive Survey on TinyML > References
“96. K. Ahmed and M. Hassan, "tinycare: A tinyml-based low-cost continuous blood pressure estimation on the extreme edge," in 2022 IEEE 10th International Conference on Healthcare Informatics (ICHI). IEEE, 2022.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ A Comprehensive Survey on TinyML > I. INTRODUCTION
“Section VI presents the remaining challenges in TinyML, open research questions, and future research directions.Finally, Section VII concludes the paper.”
§ A Comprehensive Survey on TinyML > VII. NAVIGATING THE CHALLENGES: AN OVERVIEW OF THE CHALLENGES FACING TINYML.
“## VII. NAVIGATING THE CHALLENGES: AN OVERVIEW OF THE CHALLENGES FACING TINYML.”
1 more finding
Create a free account to see every finding of this check.
Citations
Cites
DARTS: Differentiable architecture search
Liu, H., Simonyan, K., & Yang, Y. (2018) · arXiv · doi:10.48550/arxiv.1806.09055 · Cited by 3,248
A comparative study of LPWAN technologies for large-scale IoT deployment
Mekki, K., Bajic, E., Chaxel, F., & Meyer, F. (2019) · ICT Express · doi:10.1016/j.icte.2017.12.005 · Cited by 1,547 · cited 1×
“End-devices communicate with these base stations through BPSK modulation in a 100 Hz ultra-narrow band with a maximum data rate of 100 bits/s [73].”
BackgroundDeep learning-enabled medical computer vision
Esteva, A., Chou, K., Yeung, S., Naik, N., Madani, A., Mottaghi, A., Liu, Y., Topol, E., Dean, J., & Socher, R. (2021) · npj Digital Medicine · doi:10.1038/s41746-020-00376-2 · Cited by 1,406 · cited 1×
“puter vision (CV) [5], social media [6], and Autonomous”
BackgroundA survey on LPWA technology: LoRa and NB-IoT
Sinha, R. S., Wei, Y., & Hwang, S.-H. (2017) · ICT Express · doi:10.1016/j.icte.2017.03.004 · Cited by 1,020 · cited 1×
“[75] It is meant to promote long battery life and low-cost devices by utilizing one resource block of LTE networks, equivalent to 180 kHz”
BackgroundEnergy and Policy Considerations for Modern Deep Learning Research
Strubell, E., Ganesh, A., & McCallum, A. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i09.7123 · Cited by 774 · cited 2×
“Authors in [9] evaluated the estimated cost of training some benchmark Deep Learning (DL) algorithms, in terms of CO2 emissions and cloud compute time.”
BackgroundMachine Learning for Anomaly Detection: A Systematic Review
Nassif, A. B., Talib, M. A., Nasir, Q., & Dakalbab, F. M. (2021) · IEEE Access · doi:10.1109/access.2021.3083060 · Cited by 748 · cited 1×
“popular as a way to identify abnormal situations [131].”
BackgroundPerspectives in machine learning for wildlife conservation
Tuia, D., Kellenberger, B., Beery, S., Costelloe, B. R., Zuffi, S., Risse, B., Mathis, A., Mathis, M. W., van Langevelde, F., Burghardt, T., et al. (2022) · Nature Communications · doi:10.1038/s41467-022-27980-y · Cited by 740 · cited 1×
“Authors in [87] discuss the use of ML techniques, including on-animal sensors for tracking movement and bioacoustic sensors, to gather data about wildlife and the environment.”
BackgroundOn the coverage of LPWANs: range evaluation and channel attenuation model for LoRa technology
Petajajarvi, J., Mikhaylov, K., Roivainen, A., Hanninen, T., & Pettissalo, M. (2015) · 2015 14th International Conference on ITS Telecommunications (ITST) · doi:10.1109/itst.2015.7377400 · Cited by 735
A Survey on Industrial Internet of Things: A Cyber-Physical Systems Perspective
Xu, H., Yu, W., Griffith, D., & Golmie, N. (2018) · IEEE Access · doi:10.1109/access.2018.2884906 · Cited by 637
Wu, B., Wan, A., Iandola, F., Jin, P. H., & Keutzer, K. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw.2017.60 · Cited by 614
Research works printed but not linked: 20
- Soumyalatha, Study of IoT: Understanding IoT architecture, applications, issues and challenges, 478, 1
- V1.0 Results
- Dawnbench
- 2023, Microcontroller Market Size and Forecast
- Edge TPU—Run Inference at the Edge  |  Google Cloud
- 2022, Tensorflow Lite Guide
- 2022, Tensorflow Lite Micro
- 2022, Getting Started
- Situnayake, 2021, How Tensorflow Helps Edge Impulse Make ML Accessible to Embedded Engineers
- Jongboom, 2020, Introducing EON: Neural Networks in up to 55% Less Ram and 35% Less Rom
- Moreau, 2020, Announcing Fomo (Faster Objects, More Objects)
- What is LoRawan®
- Naeem, Efficient Ultra Low Power Underwater Imaging
- 2020, ElephantEdge—Hackster.Io
- 2020, Elephant Guardian—TinyMLs to Save the Giant Wild Animal
- 2017, World Population Projected to Reach 9.8 Billion in 2050, and 11.2 Billion in 2100
- Esgario, 2019, Coffee Dataset
- Oltean, 2017, Fruits 360 Dataset: A Dataset of Images Containing Fruits and Vegetables
- 2023, Plant Village
- 2022, State of the Tinyautoml Market 2022
Cited by
A Machine Learning-Oriented Survey on Tiny Machine Learning
Capogrosso, L., Cunico, F., Cheng, D. S., Fummi, F., & Cristani, M. (2024) · IEEE Access · doi:10.1109/access.2024.3365349 · Cited by 247
“It should be noted that the majority of these works are not conducted based on a well-defined and widely known Systematic Literature Review (SLR), e.g ., using Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) [44], except [36] and [43].”
BackgroundMethodHeydari, S., & Mahmoud, Q. H. (2025) · Sensors · doi:10.3390/s25103191 · Cited by 276
“Using these optimization techniques, TinyML algorithms can achieve compression ratios of up to 49 × while maintaining acceptable accuracy levels [1].”
InfluentialAdvancements in TinyML: Applications, Limitations, and Impact on IoT Devices
Elhanashi, A., Dini, P., Saponara, S., & Zheng, Q. (2024) · Electronics · doi:10.3390/electronics13173562 · Cited by 189
AI-powered biometrics for Internet of Things security: A review and future vision
Awad, A. I., Babu, A., Barka, E. S., & Shuaib, K. (2024) · Journal of Information Security and Applications · doi:10.1016/j.jisa.2024.103748 · Cited by 147
Noninvasive Diabetes Detection through Human Breath Using TinyML-Powered E-Nose
Gudiño-Ochoa, A., García-Rodríguez, J. A., Ochoa-Ornelas, R., Chávez, J. I. C., & Sánchez-Arias, D. A. (2024) · Sensors · doi:10.3390/s24041294 · Cited by 65
“The application of TensorFlow Lite or TinyML is not available for some machine learning models due to limitations in model conversion [52–54].”
Method
Metadata
- Authors
- Abadade, Y., Temouden, A., Bamoumen, H., Benamar, N., Chtouki, Y., & Hafid, A. S.
- Affiliations
- Université Ibn-Tofail (MA)
- Al Akhawayn University (MA)
- Université Moulay Ismail de Meknes (MA)
- Université de Montréal (CA)
- Venue
- IEEE Access
- Published
- 2023
- Volume
- 11
- Pages
- 96892-96922
- ISSN
- 2169-3536 (ISSN-L)
- OpenAlex
- W4383751021
- Semantic Scholar
- 0adf3bf3f8cc607221d2459d7671fc549cbc2075
- Cited by
- 366
- References
- 129
- Citations by year
- 2026: 134 · 2025: 134 · 2024: 88 · 2023: 10
- FWCI
- 40.55
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- IEEE Access (publishedVersion, cc-by-nc-nd)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
- Physical Sciences › Computer Science › Information Systems › Internet of Things and AI
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Keywords
- TinyML, resource-constrained devices, edge computing, low-power machine learning, medical applications, smart agriculture, environmental monitoring, anomaly detection, energy efficiency
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-06, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-06 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: A Comprehensive Survey on TinyML. https://app.cochat.ai/papers/10.1109/access.2023.3294111
- The paper
- Abadade, Y., Temouden, A., Bamoumen, H., Benamar, N., Chtouki, Y., & Hafid, A. S. (2023). A Comprehensive Survey on TinyML. IEEE Access. https://doi.org/10.1109/access.2023.3294111
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.