Pith. sign in

REVIEW 3 major objections 6 minor 34 references

Are Trees Really Green? A Detection Approach of IoT Malware Attacks

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Energy-aware tuning of tree-based classifiers cuts inference energy by 60-90% while keeping detection MCC near 0.60, making on-device IoT intrusion detection viable.

desk verdict Useful green-ML result undermined by test-set selection in the evaluation protocol; the energy savings are real but the 'high performance' claim is not yet established. read the letter →

arxiv 2506.07836 v1 pith:53NOPGP7 submitted 2025-06-09 cs.CR cs.AIcs.NI

Silvia Lucia Sanna , Diego Soi , Davide Maiorca , Giorgio Giacinto This is my paper
classification cs.CRcs.AIcs.NI
keywords greenmachinelearningIoTmalwaredetectionintrusionsystemhyperparameteroptimizationenergyconsumptiontree-basedclassifiersnetworktrafficanalysisMatthew'scorrelationcoefficient
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 asks whether tree-based machine-learning detectors of IoT malware can be made 'green' without sacrificing detection. It claims that optimizing the hyperparameters of decision trees, random forests, and extra-trees for a combination of Matthew's Correlation Coefficient and per-sample inference energy (in watt-hours) yields models that keep MCC around 0.60 while consuming 60-90% less energy than the default configurations. The study uses flow-level statistical features that never inspect packet payloads, so the approach is privacy-preserving and encryption-agnostic. If the claim holds, energy-aware model selection makes on-premise ML intrusion detection practical for resource-constrained IoT devices.

What carries the argument

The machinery is a two-objective hyperparameter search: a single optimizer runs dozens of trials per model, maximizing MCC and minimizing the mean energy per test sample in micro-watt-hours, with energy measured by a profiling tool that reads the processor's power meter. The 'balanced' model is selected as the Pareto-front point closest to (0,1) in the energy-MCC plane. All models are trained on flow statistics—packet timings, sizes, and TCP flag counts computed bidirectionally and per direction—excluding IP addresses, ports, and payload, which keeps the analysis privacy-preserving and independent of encryption.

What would settle it

Run the same optimization pipeline with a genuinely held-out test set that is never touched during hyperparameter selection; if MCC drops well below 0.60 or the energy savings shrink when the model faces unseen traffic, the central claim would be refuted. Alternatively, repeat the experiment on a second labeled IoT traffic dataset and check whether the balanced single-tree still reaches about 0.60 MCC with 60-90% energy reduction.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that model selection for IoT malware detection should treat test-time energy as a first-class objective rather than an afterthought. A two-objective hyperparameter search over MCC and mean watt-hours per sample, with energy measured during inference, finds a balanced configuration on the Pareto front that matches or beats the default model's MCC (around 0.60) while reducing consumed resources by 60-90%. The single decision tree is the most efficient of the three algorithms; the ensembles deliver comparable MCC but at much higher energy. A further experiment removing port-scan flows, which are easily confused with benign traffic, pushes the single-tree MCC near 0.995 at about 2.35 micro-watt-hours per sample.

Load-bearing premise

The evaluation assumes that tuning the models on the test split and then measuring them on that same split gives an unbiased estimate of real performance on unseen IoT traffic, because no separate validation set is held out.

Editorial extensions

If this is right

  • Default hyperparameter settings are far from energy-optimal for IoT detection; simply tuning for energy can cut inference cost by an order of magnitude.
  • A single decision tree with balanced tuning reaches about 0.60 MCC at roughly 8 micro-watt-hours per sample, making it a stronger on-device candidate than the ensemble models.
  • Because the features are payload-free and encryption-agnostic, the same detector can be applied to encrypted IoT traffic without decryption.
  • The near-perfect MCC after removing port-scan flows indicates that much of the remaining error is a labeling and feature-similarity problem rather than a model-capacity problem.
  • If the energy reductions transfer from the server testbed to real IoT hardware, on-premise ML-based intrusion detection becomes feasible on battery-powered devices.

Reading between the lines

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

  • The energy figures were measured on a server processor, not on an actual IoT microcontroller; real-world savings on constrained hardware may differ, and the paper's central claim depends on that transfer.
  • The paper's energy measurements depend on an energy-estimation tool that is mentioned in the text but has no corresponding entry in the reference list, so its reported accuracy is not verifiable from the paper.
  • A stricter evaluation with a separate validation set or cross-validation would tell whether the reported MCC and energy values are predictive or are partly fitted to the test split.
  • The same tuning recipe could be applied to other lightweight detectors, such as logistic regression or small neural networks, to see whether the 60-90% energy reduction is specific to tree ensembles.
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

3 major / 6 minor

Summary. This paper proposes an energy-aware hyperparameter optimization methodology for tree-based IoT malware traffic classifiers (Decision Tree, Random Forest, Extra-Trees), using Optuna with a multi-objective function that minimizes per-sample inference energy (µWh, measured with Carbon Tracker) and maximizes Matthew's Correlation Coefficient. On the Aposemat IoT-23 dataset, the authors compare default, max-green, max-MCC, and balanced models, reporting that balanced models reach MCC around 0.60 while consuming 60–90% less energy per test sample than the default models. The paper also analyzes false negatives and argues that port-scan flows resemble legitimate traffic, reporting much higher MCC after removing port-scan flows from the evaluation.

Significance. Strengths: the paper addresses a relevant and under-studied problem (energy-aware model selection for on-premise IDS); uses a public dataset; uses privacy-preserving flow statistics; the energy reduction comparison is internally consistent because the same hardware and measurement tool are used for default and optimized models; and the selected hyperparameters (e.g., shallow trees, few estimators for max-green and balanced models) provide a credible mechanistic explanation for the reductions. If the performance claims were validated on truly unseen data, the result would be practically useful for resource-constrained deployments and would complement existing green-ML literature. The main limitation is that the reported MCC values may reflect selection on the test set rather than predictive performance; the post-hoc port-scan removal also changes the task. These issues are fixable with a proper validation protocol.

major comments (3)
  1. [Section 3.2–3.3 and Table 5] Section 3.2 creates only an 80-20 train/test split, and Section 3.3 states that 'Model optimization was performed based on energy consumption and performance during the testing phase.' As written, the same 20% test split appears to be used both for Optuna model selection and for reporting the MCC and µWh values in Table 5; no separate validation set or nested cross-validation is described. The balanced-model MCC values (0.60–0.61) and the corresponding energy numbers are therefore fitted selection outcomes, not unbiased estimates on unseen traffic, and the abstract's claim that 'models maintain high performance' is not yet supported. Please report results from a held-out validation set (or an inner CV loop) that is not used during the Optuna trials.
  2. [Section 4.4 and Figure 3] The error analysis removes the entire port-scan class from the dataset, which is the largest portion of the malicious traffic (over 10.9 million of roughly 11.7 million malicious flows in Table 1), and then reports MCC near 0.995 for the balanced single-tree model on the remaining data. This is a different and much easier classification task, and it does not support the paper's detection claim for real IoT traffic, which includes port scanning. Moreover, Section 4.4 explicitly states that the authors 'did not optimize the hyperparameters... as we expected it to have low accuracy,' so the high post-removal numbers are not produced by the proposed optimization methodology. The post-hoc removal should be presented only as an illustration of dataset bias, and the abstract and conclusion should not rely on it as evidence of maintained performance.
  3. [Section 4.1 and Section 5] All experiments are performed on a server, and Section 5 concedes that no constrained device was used. The paper's concluding claim that ML-based IDS 'are suitable for running on on-premise devices' and the abstract's suggestion of suitability for resource-constrained devices go beyond the evidence, because µWh measured on an Intel i9 server cannot be transferred to low-power IoT hardware without an execution model or actual measurements. Please either temper the conclusion to state that per-inference energy is low in absolute terms, or provide measurements (or validated energy models) on constrained hardware.
minor comments (6)
  1. [Section 4.2] The text states that the default models reach 'about 99% balanced accuracy,' but Table 5 reports balanced accuracy values of approximately 94.7%; please correct the inconsistency.
  2. [Section 4.4] The sentence 'The latter is expected because the less are the flows, the less complex the model is' is not clearly correct, since µWh is reported per test sample and the hyperparameters were not re-optimized after removal; please clarify the mechanism by which removing flows reduces per-sample energy.
  3. [Table 5 and Figure 3] The axis label in Figure 3 reads 'Wh' while the text and Table 5 use 'µWh'; please make the units consistent.
  4. [General] There are several typographical errors, including 'Secion' in Section 3.3 and 'Miraii' in the Introduction; please proofread the manuscript.
  5. [References] Reference [26] (the IoT-23 dataset) is missing publication venue and year information; please complete the bibliographic entry.
  6. [Abstract and Section 4.2] The abstract refers to 'detection accuracy,' but the paper's primary metric is MCC; please align the terminology with the evaluation metrics.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported MCC values are Optuna objectives evaluated on the test split used for selection, so the 'maintain high performance' claim is a fitted result.

  1. fitted input called prediction [Section 3.2 and Section 3.3; reported in Table 5 and Section 4.3]
    "Once we built our dataset, we randomly split it into a training and a test set following an 80-20 ratio. ... Model optimization was performed based on energy consumption and performance during the testing phase. ... we asked optuna to perform 64 iterations ... The optimizations aimed to maximize the MCC and minimize the energy footprint during the inference phase."

    The same 20% split is used both for Optuna's objective (MCC and energy) and for the final Table 5 values. Selecting hyperparameters to maximize MCC and minimize energy on the test split, then reporting the MCC and energy of the selected model on that same split, makes the reported values selection optima rather than unbiased estimates for unseen traffic. The balanced model is specifically chosen as the point geometrically closest to (0,1) in that test-set Pareto front, so its headline 0.60 MCC and 60-90% energy reduction are constructed by the selection rule.

full rationale

The paper's energy-reduction comparison is credible and self-contained: default and optimized configurations are measured on the same hardware, and the direction of the effect is plausible from the chosen hyperparameters. However, the performance half of the central claim is statistically forced. Section 3.2 creates only an 80-20 train/test split, and Section 3.3 states that optimization is performed 'based on energy consumption and performance during the testing phase.' No separate validation set is described, so the test split is used both to select hyperparameters and to compute the reported MCC values. The balanced model is then selected as the Pareto-front point closest to (0,1), meaning the reported MCC and energy are the very objectives being optimized. This is a fitted-input-called-prediction pattern for the performance claim, not a circularity based on self-citation; the paper contains no load-bearing self-citation chain or imported uniqueness theorem. Weighing the independent energy result against the non-independent MCC result, the overall circularity score is 6: one of the paper's two headline predictions reduces by construction, while the other retains independent content.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or conceptual entities; it applies existing models, tools, and datasets. The main burden lies in fitted hyperparameters, the representativeness of the single dataset, the accuracy of the energy estimator, and the post-hoc removal of port-scan flows.

free parameters (3)
  • single-tree balanced hyperparameters = max_depth=13, min_leaf=5, min_split=13
    Selected by Optuna to maximize MCC and minimize energy on the same data used for evaluation; these values directly determine the reported balanced model results.
  • random forest balanced hyperparameters = max_depth=17, min_leaf=6, min_split=20, max_features=7, estimators=18
    Fitted by Optuna to the evaluation data; controls the reported energy and MCC trade-off for the random forest.
  • extra-trees balanced hyperparameters = max_depth=14, min_leaf=2, min_split=18, max_features=24, estimators=204
    Fitted by Optuna to the evaluation data; controls the reported energy and MCC trade-off for the extra-trees model.
assumptions (4)
  • domain assumption The IoT-23 dataset is representative of realistic IoT malware traffic.
    Section 3.1 uses this dataset exclusively and the abstract generalizes to 'common network cyberattacks in different IoT scenarios' without external validation.
  • domain assumption Carbon Tracker's RAPL-based estimates of CPU and RAM energy are accurate enough to compare model variants.
    Section 2.1 and Section 3.3 rely on this tool for all µWh measurements without independent ground-truth validation in this paper.
  • domain assumption Energy measured on the server used for training reflects the energy a constrained IoT device would consume.
    Section 4.1 states experiments used a server to ensure reproducibility, and the authors list the lack of a constrained device as a limitation in Section 5.
  • ad hoc to paper Port-scan flows may be removed for error analysis because they resemble legitimate flows.
    Section 4.4 drops the largest attack class and then reports near-perfect MCC; this exclusion is specific to this paper's analysis and changes the detection problem.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Are Trees Really Green? A Detection Approach of IoT Malware Attacks." pith.science (2026). https://pith.science/paper/53NOPGP7

@misc{pith2026250607836,
  author       = {Pith},
  title        = {Pith review of: Are Trees Really Green? A Detection Approach of IoT Malware Attacks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/53NOPGP7}},
  note         = {Machine review of arXiv:2506.07836}
}
read the original abstract

Nowadays, the Internet of Things (IoT) is widely employed, and its usage is growing exponentially because it facilitates remote monitoring, predictive maintenance, and data-driven decision making, especially in the healthcare and industrial sectors. However, IoT devices remain vulnerable due to their resource constraints and difficulty in applying security patches. Consequently, various cybersecurity attacks are reported daily, such as Denial of Service, particularly in IoT-driven solutions. Most attack detection methodologies are based on Machine Learning (ML) techniques, which can detect attack patterns. However, the focus is more on identification rather than considering the impact of ML algorithms on computational resources. This paper proposes a green methodology to identify IoT malware networking attacks based on flow privacy-preserving statistical features. In particular, the hyperparameters of three tree-based models -- Decision Trees, Random Forest and Extra-Trees -- are optimized based on energy consumption and test-time performance in terms of Matthew's Correlation Coefficient. Our results show that models maintain high performance and detection accuracy while consistently reducing power usage in terms of watt-hours (Wh). This suggests that on-premise ML-based Intrusion Detection Systems are suitable for IoT and other resource-constrained devices.

Figures

Figures reproduced from arXiv: 2506.07836 by the authors.

Figure 1
Figure 1. Training workflow with the basic blocks of the methodology: dataset la [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Confusion Matrix and Pareto front for the single-tree models showing [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Comparison between performance of the optimized single-tree models [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Pareto front for the tree models after removing port-scanning flows. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

  1. [1]

    What is the mirai botnet? https://www.cloudflare.com/it- it/learning/ddos/glossary/mirai-botnet/, 2025

    work page 2025

  2. [2]

    The internet of things: A survey

    Luigi Atzori, Antonio Iera, and Giacomo Morabito. The internet of things: A survey. Computer Networks, 54(15):2787–2805, 2010

    work page 2010

  3. [3]

    Maldist: From encrypted traffic classification to malware traffic detection and classification

    Ofek Bader, Adi Lichy, Chen Hajaj, Ran Dubin, and Amit Dvir. Maldist: From encrypted traffic classification to malware traffic detection and classification. In 2022 IEEE 19th Annual Consumer Communications & Networking Conference (CCNC), pages 527–533, 2022

    work page 2022

  4. [4]

    Extending c2 traffic detection methodologies: From tls 1.2 to tls 1.3-enabled mal- ware

    Diogo Barradas, Carlos Novo, Bernardo Portela, Sofia Romeiro, and Nuno Santos. Extending c2 traffic detection methodologies: From tls 1.2 to tls 1.3-enabled mal- ware. InProceedings of the 27th International Symposium on Research in Attacks, Intrusions and Defenses, RAID ’24, New York, NY, USA, 2024. Association for Computing Machinery

    work page 2024

  5. [5]

    Disclosure: detecting botnet command and control servers through large-scale netflow analysis

    Leyla Bilge, Davide Balzarotti, William Robertson, Engin Kirda, and Christo- pher Kruegel. Disclosure: detecting botnet command and control servers through large-scale netflow analysis. InProceedings of the 28th Annual Computer Security Applications Conference, ACSAC ’12, page 129–138, New York, NY, USA, 2012. Association for Computing Machinery

    work page 2012

  6. [6]

    eco2ai:Carbonemissionstrack- ing of machine learning models as the first step towards sustainable ai.Doklady Mathematics, 2022

    Semen Budennyy, Vladimir Lazarev, Nikita Zakharenko, Alexey Korovin, Olga Plosskaya, Denis Dimitrov, Vladimir Arkhipkin, Ivan Oseledets, Ivan Barsola, Ilya Egorov,AleksandraKosterina,andLeonidZhukov. eco2ai:Carbonemissionstrack- ing of machine learning models as the first step towards sustainable ai.Doklady Mathematics, 2022

    work page 2022

  7. [7]

    ’hide ’n seek’ botnet uses peer-to-peer infrastructure to compro- mise iot devices

    Trend Business. ’hide ’n seek’ botnet uses peer-to-peer infrastructure to compro- mise iot devices. https://www.trendmicro.com/vinfo/us/security/news/internet- of-things/-hide-n-seek-botnet-uses-peer-to-peer-infrastructure-to-compromise-iot- devices, 2018

    work page 2018

  8. [8]

    Encryption-agnostic classifiers of traffic originators and their applica- tion to anomaly detection.Computers & Electrical Engineering, 97:107621, 2022

    Daniele Canavese, Leonardo Regano, Cataldo Basile, Gabriele Ciravegna, and An- tonio Lioy. Encryption-agnostic classifiers of traffic originators and their applica- tion to anomaly detection.Computers & Electrical Engineering, 97:107621, 2022

    work page 2022

Show all 34 references
  1. [9]

    Fairandgreenhyperpa- rameter optimization via multi-objective and multiple information source bayesian optimization

    AntonioCandelieri,AndreaPonti,andFrancescoArchetti. Fairandgreenhyperpa- rameter optimization via multi-objective and multiple information source bayesian optimization. Machine Learning, 113, 2024

    2024

  2. [10]

    Buchanan

    Andrew Churcher, Rehmat Ullah, Jawad Ahmad, Sadaqat ur Rehman, Fawad Ma- sood, Mandar Gogate, Fehaid Alqahtani, Boubakr Nour, and William J. Buchanan. An experimental analysis of attack classification using machine learning in iot net- works. Sensors, 21(2), 2021

    2021

  3. [11]

    C2miner: Tricking iot malware into revealing live command & control servers

    Ali Davanian, Michail Faloutsos, and Martina Lindorfer. C2miner: Tricking iot malware into revealing live command & control servers. InProceedings of the 19th ACM Asia Conference on Computer and Communications Security, ASIA CCS ’24, New York, NY, USA, 2024. Association for Co...

    2024

  4. [12]

    How to measure energy consumption in machine learning algo- rithms

    Eva García-Martín, Niklas Lavesson, Håkan Grahn, Emiliano Casalicchio, and Veselka Boeva. How to measure energy consumption in machine learning algo- rithms. In ECML PKDD 2018 Workshops, pages 243–255, Cham, 2019. Springer International Publishing

    2018

  5. [13]

    Estimation of energy consumption in machine learning.Journal of Parallel and Distributed Computing, 134:75–88, 2019

    Eva García-Martín, Crefeda Faviola Rodrigues, Graham Riley, and Håkan Grahn. Estimation of energy consumption in machine learning.Journal of Parallel and Distributed Computing, 134:75–88, 2019. Green Machine Learning for IoT Detection 17

    2019

  6. [14]

    Extremely randomized trees

    Pierre Geurst, Damien Ernst, and Louis Wehenkel. Extremely randomized trees. Machine Learning, 63, 2006

    2006

  7. [15]

    Long, Bui D

    Truong Thu Huong, Ta Phuong Bac, Dao M. Long, Bui D. Thang, Nguyen T. Binh, Tran D. Luong, and Tran Kim Phuc. Lockedge: Low-complexity cyberattack detection in iot edge computing.IEEE Access, 9:29696–29710, 2021

    2021

  8. [16]

    Iacovos Ioannou, Prabagarane Nagaradjane, Pelin Angin, Palaniappan Balasub- ramanian, Karthick Jeyagopal Kavitha, Palani Murugan, and Vasos Vassiliou. Gemlids-miot: A green effective machine learning intrusion detection system based on federated learning for medical iot networ...

    2024

  9. [17]

    Flow- guard: An intelligent edge defense mechanism against iot ddos attacks

    Yizhen Jia, Fangtian Zhong, Arwa Alrawais, Bei Gong, and Xiuzhen Cheng. Flow- guard: An intelligent edge defense mechanism against iot ddos attacks. IEEE Internet of Things Journal, 7(10):9552–9562, 2020

    2020

  10. [18]

    Iot network traffic classification using machine learning algorithms: An experimental analysis

    Rakesh Kumar, Mayank Swarnkar, Gaurav Singal, and Neeraj Kumar. Iot network traffic classification using machine learning algorithms: An experimental analysis. IEEE Internet of Things Journal, 9(2):989–1008, 2022

    2022

  11. [19]

    New machine learning algorithm: Ran- dom forest

    Yanli Liu, Yourong Wang, and Jian Zhang. New machine learning algorithm: Ran- dom forest. In Baoxiang Liu, Maode Ma, and Jincai Chang, editors,Information Computing and Applications, pages 246–252, Berlin, Heidelberg, 2012. Springer Berlin Heidelberg

    2012

  12. [20]

    Keertikumar M., Shubham M., and R.M. Banakar. Evolution of iot in smart vehicles: An overview. In2015 International Conference on Green Computing and Internet of Things (ICGCIoT), pages 804–809, 2015

    2015

  13. [21]

    Iot: Internet of threats? a survey of practical security vulnerabilities in real iot devices.IEEE Internet of Things Journal, 6(5):8182–8201, 2019

    Francesca Meneghello, Matteo Calore, Daniel Zucchetto, Michele Polese, and An- drea Zanella. Iot: Internet of threats? a survey of practical security vulnerabilities in real iot devices.IEEE Internet of Things Journal, 6(5):8182–8201, 2019

    2019

  14. [22]

    Threats of internet-of- thing on environmental sustainability by e-waste.Sustainability, 2022

    Fathi Batoul Modarress, Alexander Ansari, and Ansari Al. Threats of internet-of- thing on environmental sustainability by e-waste.Sustainability, 2022

    2022

  15. [23]

    Jurcut, and Mari- anne A

    Basem Ibrahim Mukhtar, Mahmoud Said Elsayed, Anca D. Jurcut, and Mari- anne A. Azer. Iot vulnerabilities and attacks: Silex malware case study.Symmetry, 15(11), 2023

    2023

  16. [24]

    Overview of use of decision tree algorithms in machine learning

    ArundhatiNavada,AamirNizamAnsari,SiddharthPatil,andBalwantA.Sonkam- ble. Overview of use of decision tree algorithms in machine learning. In2011 IEEE Control and System Graduate Research Colloquium, pages 37–42, 2011

    2011

  17. [25]

    Landscape of iot security.Computer Science Review, 44:100467, 2022

    Eryk Schiller, Andy Aidoo, Jara Fuhrer, Jonathan Stahl, Michael Ziörjen, and Burkhard Stiller. Landscape of iot security.Computer Science Review, 44:100467, 2022

    2022

  18. [26]

    Iot-23: A labeled dataset with malicious and benign iot network traffic

    Maria Jose Erquiaga Sebastian Garcia, Agustin Parmisano. Iot-23: A labeled dataset with malicious and benign iot network traffic

  19. [27]

    Selcuk Uluagac, and Vehbi Cagri Gun- gor

    Nazli Tekin, Abbas Acar, Ahmet Aris, A. Selcuk Uluagac, and Vehbi Cagri Gun- gor. Energy consumption of on-device machine learning models for iot intrusion detection. Internet of Things, 21:100670, 2023

    2023

  20. [28]

    Charles Edison Tripp, Jordan Perr-Sauer, Jamil Gafur, Amabarish Nag, Avi Purkayastha, Sagi Zisman, and Erik A. Bensen. Measuring the energy consumption and efficiency of deep neural networks: An empirical analysis and design recom- mendations, 2024

    2024

  21. [29]

    Visualization of pareto front approximations in evo- lutionary multiobjective optimization: A critical review and the prosection method

    Tea Tušar and Bogdan Filipič. Visualization of pareto front approximations in evo- lutionary multiobjective optimization: A critical review and the prosection method. IEEE Transactions on Evolutionary Computation, 19(2):225–245, 2015. 18 Sanna et al

    2015

  22. [30]

    ’torii’ breaks new ground for iot malware

    Jai Vijayan. ’torii’ breaks new ground for iot malware. https://www.darkreading.com/cyberattacks-data-breaches/-torii-breaks-new- ground-for-iot-malware, 2018

    2018

  23. [31]

    An evolutionary study of iot malware.IEEE Internet of Things Journal, 8(20):15422– 15440, 2021

    Huanran Wang, Weizhe Zhang, Hui He, Peng Liu, Daniel Xiapu Luo, Yang Liu, Jiawei Jiang, Yan Li, Xing Zhang, Wenmao Liu, Runzi Zhang, and Xing Lan. An evolutionary study of iot malware.IEEE Internet of Things Journal, 8(20):15422– 15440, 2021

    2021

  24. [32]

    Internet of things in industries: A survey

    Li Da Xu, Wu He, and Shancang Li. Internet of things in industries: A survey. IEEE Transactions on Industrial Informatics, 10(4):2233–2243, 2014

    2014

  25. [33]

    Uncov- ering energy-efficient practices in deep learning training: Preliminary steps towards green ai

    Tim Yarally, Luıs Cruz, Daniel Feitosa, June Sallou, and Arie van Deursen. Uncov- ering energy-efficient practices in deep learning training: Preliminary steps towards green ai. In2023 IEEE/ACM 2nd International Conference on AI Engineering – Software Engineering for AI (CAIN)...

    2023

  26. [34]

    Yokoyama, Mariza Ferro, and Bruno Schulze

    André M. Yokoyama, Mariza Ferro, and Bruno Schulze. A multi-objective hyper- parameter optimization for machine learning using genetic algorithms: A green ai centric approach. In Ana Cristina Bicharra Garcia, Mariza Ferro, and Julio Cesar Rodríguez Ribón, editors,Advances in A...

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.