arxiv: 2604.07953 · v1 · submitted 2026-04-09 · 💻 cs.LG · cs.AI

Recognition: no theorem link

Pruning Extensions and Efficiency Trade-Offs for Sustainable Time Series Classification

Raphael Fischer , Angus Dempster , Sebastian Buschj\"ager , Matthias Jakobs , Urav Maniar , Geoffrey I. Webb

Authors on Pith no claims yet

Pith reviewed 2026-05-10 18:11 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords time series classificationmodel pruningenergy efficiencyhybrid classifierssustainable machine learningHydraQuantHydrant

0 comments

The pith

A theoretically bounded pruning method applied to hybrid time series classifiers cuts energy use by up to 80 percent with under 5 percent accuracy loss.

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

The paper builds a holistic framework to measure both predictive performance and energy consumption in time series classification across many datasets and hardware setups. It applies a pruning technique with theoretical guarantees to established hybrid models and introduces Hydrant as a new prunable combination of them. Experiments covering thousands of configurations show that pruning delivers large efficiency gains while keeping accuracy nearly intact. This shifts focus in the field from accuracy alone toward practical resource use and sustainability.

Core claim

A theoretically bounded pruning strategy can be extended to leading hybrid time series classifiers including Hydra and Quant, as well as to the new combined model Hydrant. When applied across 20 MONSTER datasets and three hardware platforms, the approach reduces energy consumption by up to 80 percent while incurring less than 5 percent accuracy loss in most cases, advancing reproducible and energy-aware practice in time series classification.

What carries the argument

The theoretically bounded pruning strategy, which removes parts of hybrid classifier architectures while preserving most predictive performance.

If this is right

Hybrid TSC models can be made substantially more energy efficient through bounded pruning without major performance penalties.
A single pruning technique works across multiple related classifier families and compute environments.
Sustainable TSC practice requires joint reporting of accuracy and energy metrics rather than accuracy alone.
The Hydrant architecture provides a flexible base for further efficiency improvements in time series tasks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar pruning ideas could be tested on non-hybrid time series models or other sequential data tasks to check wider applicability.
Energy measurements on additional hardware types might reveal whether the 80 percent savings hold under different power constraints.
Integrating the pruning step into automated model selection pipelines could help practitioners balance accuracy and power use more easily.

Load-bearing premise

The pruning method transfers to the tested hybrid models and hardware without hidden costs or dataset-specific biases that would change the reported energy-accuracy trade-offs.

What would settle it

Running the same pruned models on a fresh set of time series datasets or hardware where energy savings exceed 5 percent accuracy drop or produce unexpected runtime overheads would falsify the main result.

Figures

Figures reproduced from arXiv: 2604.07953 by Angus Dempster, Geoffrey I. Webb, Matthias Jakobs, Raphael Fischer, Sebastian Buschj\"ager, Urav Maniar.

**Figure 1.** Figure 1: Energy (x) versus accuracy (y) performance of TS classifiers when deployed for inference on two processors (left / right), demonstrating intricate trade-offs [18]. While many claimed efficiency improvements for TSC [41, 8, 21, 10, 11], we observe a central problem—the stateof-the-art experiments are not unified with regard to evaluation configurations and criteria. Most works overly focus on predictive qu… view at source ↗

**Figure 2.** Figure 2: Performance statistics and trade-offs among TSC models ( [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Figure 3: Critical differences [12] among original and pruned hybrids, showing how energy draw and compound [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

**Figure 4.** Figure 4: Pruning rate ablation study on model quality (left) and linearly decreasing energy demand (right), averaged [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

**Figure 5.** Figure 5: Impacts on model energy draw across all datasets when using the [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

read the original abstract

Time series classification (TSC) enables important use cases, however lacks a unified understanding of performance trade-offs across models, datasets, and hardware. While resource awareness has grown in the field, TSC methods have not yet been rigorously evaluated for energy efficiency. This paper introduces a holistic evaluation framework that explicitly explores the balance of predictive performance and resource consumption in TSC. To boost efficiency, we apply a theoretically bounded pruning strategy to leading hybrid classifiers - Hydra and Quant - and present Hydrant, a novel, prunable combination of both. With over 4000 experimental configurations across 20 MONSTER datasets, 13 methods, and three compute setups, we systematically analyze how model design, hyperparameters, and hardware choices affect practical TSC performance. Our results showcase that pruning can significantly reduce energy consumption by up to 80% while maintaining competitive predictive quality, usually costing the model less than 5% of accuracy. The proposed methodology, experimental results, and accompanying software advance TSC toward sustainable and reproducible practice.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Pruning cuts energy use substantially in these TSC models but the savings hinge on clean measurement of actual hardware draw with no unaccounted overhead.

read the letter

The main thing to know is that the authors show pruning can drop energy consumption by up to 80 percent in hybrid time series classifiers while keeping accuracy loss under 5 percent in most cases, and they back it with thousands of runs across datasets and hardware. They also introduce Hydrant as a prunable mix of Hydra and Quant plus a framework that scores both accuracy and resource use together instead of treating them separately. That combination is the concrete addition here. The large experiment count stands out as useful: 20 MONSTER datasets, 13 methods, three compute setups, and over 4000 configurations give a broad picture of how model choice, hyperparameters, and hardware interact on the efficiency side. The accompanying software mention is a plus for anyone who wants to reproduce or extend the trade-off curves. The soft spots sit mainly in the measurement details. The headline savings rest on the pruning strategy transferring cleanly to these hybrids without extra runtime costs or instrumentation biases on the three hardware platforms. If the energy figures come only from parameter counts or FLOPs rather than wall-clock power draw, or if pruning code itself adds measurable overhead, the 80 percent number becomes harder to trust in real deployments. The abstract does not spell out error bars, statistical tests, or exact exclusion rules for the datasets, so the consistency of the under-5-percent accuracy claim needs checking in the full methods. This work is aimed at researchers who deploy time series models on power-limited devices or who want practical benchmarks for sustainable ML. A reader looking for concrete efficiency numbers and an evaluation template will find value even if they end up adjusting the pruning bounds for their own hardware. It deserves a serious referee because the scale is large enough to be informative and the claims are falsifiable with the right measurement protocol. I would send it out for review rather than desk reject, with the expectation that the methods section will need tightening on energy accounting.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces a holistic evaluation framework for time series classification (TSC) that balances predictive performance against energy consumption across models, datasets, and hardware. It applies a theoretically bounded pruning strategy to the hybrid classifiers Hydra and Quant, introduces the new prunable model Hydrant as their combination, and reports results from over 4000 experimental configurations on 20 MONSTER datasets using 13 methods and three compute setups. The central empirical claim is that this pruning reduces energy consumption by up to 80% while incurring less than 5% accuracy loss in most cases, supported by accompanying open software for reproducibility.

Significance. If the energy-accuracy trade-offs hold under rigorous measurement, the work provides a valuable empirical foundation for sustainable TSC practice, including a new hybrid architecture and extensive cross-hardware comparisons that go beyond typical accuracy-only benchmarks. The scale of the experimental campaign (4000+ configurations) and release of software are clear strengths that could enable follow-on reproducible research in resource-aware time series methods.

major comments (2)

[§4 and §5] §4 (Experimental Setup) and §5 (Results): The claim that pruning transfers to Hydra, Quant, and Hydrant without hidden overheads requires explicit confirmation that wall-clock energy measurements on the three hardware platforms account for any additional runtime costs introduced by the pruning implementation itself; if the theoretical bound only constrains parameter count or FLOPs (rather than measured joules), the reported 80% savings may not generalize.
[Table 2] Table 2 (or equivalent accuracy-energy summary table): The statement that accuracy cost is 'usually less than 5%' across 4000+ configurations needs accompanying statistical tests (e.g., paired t-tests or confidence intervals per dataset/hardware) rather than aggregate reporting, as dataset-specific biases could undermine the cross-setup generalization.

minor comments (2)

[§3] The abstract and §3 should clarify the exact definition of the 'theoretically bounded pruning strategy' (e.g., which norm or sparsity constraint is used) to allow readers to assess its applicability to the hybrid architectures without re-deriving it.
[Figure 3] Figure 3 (energy vs. accuracy plots): Axis scales and legend clarity could be improved to distinguish the three hardware setups more distinctly, especially when overlaying pruned vs. unpruned variants.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback, which has helped us improve the clarity and rigor of our work. We provide detailed responses to each major comment below.

read point-by-point responses

Referee: [§4 and §5] §4 (Experimental Setup) and §5 (Results): The claim that pruning transfers to Hydra, Quant, and Hydrant without hidden overheads requires explicit confirmation that wall-clock energy measurements on the three hardware platforms account for any additional runtime costs introduced by the pruning implementation itself; if the theoretical bound only constrains parameter count or FLOPs (rather than measured joules), the reported 80% savings may not generalize.

Authors: We appreciate this point. Our energy consumption figures are derived from direct measurements of wall-clock execution time multiplied by the hardware's power consumption on each of the three platforms. These real-world measurements necessarily incorporate any additional runtime costs associated with the pruning implementation. The theoretical bounds on pruning are used to determine safe pruning ratios, but the savings claims are based on empirical joule estimates, not theoretical proxies. We will add an explicit statement in Section 4 to confirm this and describe the measurement methodology in more detail. revision: yes
Referee: [Table 2] Table 2 (or equivalent accuracy-energy summary table): The statement that accuracy cost is 'usually less than 5%' across 4000+ configurations needs accompanying statistical tests (e.g., paired t-tests or confidence intervals per dataset/hardware) rather than aggregate reporting, as dataset-specific biases could undermine the cross-setup generalization.

Authors: The referee is correct that aggregate statistics can obscure important variations. To address this, we will revise the presentation of results to include statistical tests. Specifically, we plan to report 95% confidence intervals for the accuracy loss per dataset and hardware configuration, along with paired t-test results comparing pruned and unpruned models. This will provide stronger evidence that the accuracy cost is generally below 5% and help identify any outlier cases. The revision will be incorporated in the updated manuscript. revision: yes

Circularity Check

0 steps flagged

No circularity: claims rest on empirical measurements, not self-referential derivations

full rationale

The paper presents a holistic experimental framework evaluating pruning on hybrid TSC models (Hydra, Quant, Hydrant) across 4000+ configurations on 20 datasets and three hardware setups. Central results (up to 80% energy reduction with <5% accuracy cost) are reported as direct experimental outcomes rather than predictions derived from equations or parameters fitted within the same study. No self-definitional steps, fitted-input predictions, or load-bearing self-citation chains appear in the derivation of the main claims; the theoretically bounded pruning is applied as an external strategy and validated empirically. The work is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract does not identify any free parameters, axioms, or invented entities. The pruning strategy is stated to be theoretically bounded, but no explicit assumptions or additional postulated constructs are described.

pith-pipeline@v0.9.0 · 5485 in / 1234 out tokens · 46308 ms · 2026-05-10T18:11:45.554993+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

41 extracted references · 34 canonical work pages

[1]

Green AI: a Preliminary Empirical Study on Energy Consumption in DL Models Across Different Runtime Infrastructures

Negar Alizadeh and Fernando Castor. “Green AI: a Preliminary Empirical Study on Energy Consumption in DL Models Across Different Runtime Infrastructures”. In:International Conference on AI Engineering - Software Engineering for AI. Lisbon, Portugal, 2024, pp. 134–139.ISBN: 9798400705915.DOI: 10.1145/3644815. 3644967

work page doi:10.1145/3644815 2024
[2]

Joint Leaf-Refinement and Ensemble Pruning through L1 Regu- larization

Sebastian Buschjäger and Katharina Morik. “Joint Leaf-Refinement and Ensemble Pruning through L1 Regu- larization”. In:Data Mining and Knowledge Discovery37.3 (2023), pp. 1230–1261.ISSN: 1573-756X.DOI: 10.1007/s10618-023-00921-z

work page doi:10.1007/s10618-023-00921-z 2023
[3]

Realization of Random Forest for Real-Time Evaluation through Tree Framing

Sebastian Buschjäger et al. “Realization of Random Forest for Real-Time Evaluation through Tree Framing”. In: International Conference on Data Mining. 2018.DOI:10.1109/ICDM.2018.00017

work page doi:10.1109/icdm.2018.00017 2018
[4]

Bottom-up and top-down: Predicting personality with psycholinguistic and language model features,

Nestor Cabello et al. “Fast and Accurate Time Series Classification Through Supervised Interval Search”. In: International Conference on Data Mining. 2020, pp. 948–953.DOI:10.1109/ICDM50108.2020.00107

work page doi:10.1109/icdm50108.2020.00107 2020
[5]

POCKET: Pruning Random Convolution Kernels for Time Series Classification from a Feature Selection Perspective

Shaowu Chen, Weize Sun, Lei Huang, et al. “POCKET: Pruning Random Convolution Kernels for Time Series Classification from a Feature Selection Perspective”. In:Knowledge-Based Systems300 (2024).DOI: 10.1016/j.knosys.2024.112253

work page doi:10.1016/j.knosys.2024.112253 2024
[6]

MONSTER: Monash Scalable Time Series Evaluation Repository

Angus Dempster, Navid Mohammadi Foumani, Chang Wei Tan, et al. “MONSTER: Monash Scalable Time Series Evaluation Repository”. In:Journal of Data-Centric Machine Learning Research2 (15 2025).URL: https://openreview.net/forum?id=XauSqSfZfc

2025
[7]

ROCKET: exceptionally fast and accurate time series classification using random convolutional kernels

Angus Dempster, François Petitjean, and Geoffrey I. Webb. “ROCKET: exceptionally fast and accurate time series classification using random convolutional kernels”. In:Data Mining and Knowledge Discovery34.5 (2020), pp. 1454–1495.ISSN: 1573-756X.DOI:10.1007/s10618-020-00701-z

work page doi:10.1007/s10618-020-00701-z 2020
[8]

Hydra: competing convolutional kernels for fast and accurate time series classification

Angus Dempster, Daniel F. Schmidt, and Geoffrey I. Webb. “Hydra: competing convolutional kernels for fast and accurate time series classification”. In:Data Mining and Knowledge Discovery37.5 (2023), pp. 1779–1805. ISSN: 1573-756X.DOI:10.1007/s10618-023-00939-3

work page doi:10.1007/s10618-023-00939-3 2023
[9]

MiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series Classification

Angus Dempster, Daniel F. Schmidt, and Geoffrey I. Webb. “MiniRocket: A Very Fast (Almost) Deterministic Transform for Time Series Classification”. In:Conference on Knowledge Discovery & Data Mining. 2021, pp. 248–257.ISBN: 978-1-4503-8332-5.DOI:10.1145/3447548.3467231

work page doi:10.1145/3447548.3467231 2021
[10]

quant: a minimalist interval method for time series classification

Angus Dempster, Daniel F. Schmidt, and Geoffrey I. Webb. “quant: a minimalist interval method for time series classification”. In:Data Mining and Knowledge Discovery38.4 (2024), pp. 2377–2402.ISSN: 1573-756X.DOI: 10.1007/s10618-024-01036-9

work page doi:10.1007/s10618-024-01036-9 2024
[11]

Highly Scalable Time Series Classification for Very Large Datasets

Angus Dempster, Chang Wei Tan, Lynn Miller, et al. “Highly Scalable Time Series Classification for Very Large Datasets”. In:Advanced Analytics and Learning on Temporal Data. 2025, pp. 80–95.ISBN: 978-3-031-77066-1. DOI:10.1007/978-3-031-77066-1_5

work page doi:10.1007/978-3-031-77066-1_5 2025
[12]

Statistical Comparisons of Classifiers over Multiple Data Sets

Janez Demšar. “Statistical Comparisons of Classifiers over Multiple Data Sets”. In:Journal of Machine Learning Research7.1 (2006), pp. 1–30.URL:http://jmlr.org/papers/v7/demsar06a.html

2006
[13]

Beyond Sleep Staging: Advancing End-to-End Event Scoring in Sleep Medicine

Sarah Dietz-Terjung, Matthias Jakobs, Chiara Labeit, et al. “Beyond Sleep Staging: Advancing End-to-End Event Scoring in Sleep Medicine”. DE. In:Pneumologie79.S 01 (2025).ISSN: 0934-8387.DOI: 10.1055/s- 0045-1804722

work page doi:10.1055/s- 2025
[14]

Time Series Classification: A review of Algorithms and Implementations

Johann Faouzi. “Time Series Classification: A review of Algorithms and Implementations”. In:Time Series Analysis - Recent Advances, New Perspectives and Applications. IntechOpen, 2024, p. 298.DOI: 10.5772/ intechopen.1004810. (Visited on 01/26/2026)

2024
[15]

Deep learning for time series classification: a review

Hassan Ismail Fawaz, Germain Forestier, Jonathan Weber, et al. “Deep learning for time series classification: a review”. In:Data Mining and Knowledge Discovery33.4 (2019), pp. 917–963.DOI: 10.1007/s10618-019- 00619-1

work page doi:10.1007/s10618-019- 2019
[16]

Advancing the Sustainability of Machine Learning and Artificial Intelligence via Labeling and Meta-Learning

Raphael Fischer. “Advancing the Sustainability of Machine Learning and Artificial Intelligence via Labeling and Meta-Learning”. PhD thesis. TU Dortmund University, 2025.DOI:10.17877/DE290R-25716

work page doi:10.17877/de290r-25716 2025
[17]

Ground-Truthing

Raphael Fischer.Ground-Truthing AI Energy Consumption: Validating CodeCarbon Against External Measure- ments. 2025.DOI:10.48550/arXiv.2509.22092. 10 PRUNINGEXTENSIONS ANDEFFICIENCYTRADE-OFFS FORSUSTAINABLETIMESERIESCLASSIFICATION

work page doi:10.48550/arxiv.2509.22092 2025
[18]

Towards More Sustainable and Trustworthy Reporting in Machine Learning

Raphael Fischer, Thomas Liebig, and Katharina Morik. “Towards More Sustainable and Trustworthy Reporting in Machine Learning”. In:Data Mining and Knowledge Discovery(2024).ISSN: 1573-756X.DOI: 10.1007/ s10618-024-01020-3

2024
[19]

AutoXPCR: Automated Multi-Objective Model Selection for Time Series Forecasting

Raphael Fischer and Amal Saadallah. “AutoXPCR: Automated Multi-Objective Model Selection for Time Series Forecasting”. In:Conference on Knowledge Discovery and Data Mining. 2024, pp. 806–815.ISBN: 979-8-4007-0490-1.DOI:10.1145/3637528.3672057

work page doi:10.1145/3637528.3672057 2024
[20]

No Cloud on the Horizon: Probabilistic Gap Filling in Satellite Image Series

Raphael Fischer et al. “No Cloud on the Horizon: Probabilistic Gap Filling in Satellite Image Series”. In: International Conference on Data Science and Advanced Analytics (DSAA). 2020, pp. 546–555.DOI: 10.1109/ DSAA49011.2020.00069

work page arXiv 2020
[21]

Improving position encoding of transformers for multivariate time series classification

Navid Mohammadi Foumani et al. “Improving position encoding of transformers for multivariate time series classification”. In:Data Mining and Knowledge Discovery38.1 (2024), pp. 22–48.ISSN: 1573-756X.DOI: 10.1007/s10618-023-00948-2

work page doi:10.1007/s10618-023-00948-2 2024
[22]

InceptionTime: Finding AlexNet for time series classification

Hassan Ismail Fawaz, Benjamin Lucas, Germain Forestier, et al. “InceptionTime: Finding AlexNet for time series classification”. en. In:Data Mining and Knowledge Discovery34.6 (2020), pp. 1936–1962.ISSN: 1573-756X. DOI:10.1007/s10618-020-00710-y. (Visited on 01/26/2026)

work page doi:10.1007/s10618-020-00710-y 2020
[23]

LSTM Fully Convolutional Networks for Time Series Classification

Fazle Karim et al. “LSTM Fully Convolutional Networks for Time Series Classification”. In:IEEE Access6 (2018), pp. 1662–1669.ISSN: 2169-3536.DOI:10.1109/access.2017.2779939

work page doi:10.1109/access.2017.2779939 2018
[24]

Lift what you can: green online learning with heterogeneous ensembles

Kirsten Köbschall et al. “Lift what you can: green online learning with heterogeneous ensembles”. In:Data Mining and Knowledge Discovery40.3 (Mar. 2026), p. 32.ISSN: 1573-756X.DOI: 10.1007/s10618-026- 01200-3.URL:https://doi.org/10.1007/s10618-026-01200-3

work page doi:10.1007/s10618-026- 2026
[25]

Pruning vs Quantization: Which is Better?

Andrey Kuzmin et al. “Pruning vs Quantization: Which is Better?” In:Advances in Neural Information Processing Systems. V ol. 36. 2023, pp. 62414–62427.URL:https://dl.acm.org/doi/10.5555/3666122.3668847

work page doi:10.5555/3666122.3668847 2023
[26]

2019.DOI:10.48550/arXiv.1909.07872

Markus Löning, Anthony Bagnall, Sajaysurya Ganesh, et al.sktime: A Unified Interface for Machine Learning with Time Series. 2019.DOI:10.48550/arXiv.1909.07872

work page doi:10.48550/arxiv.1909.07872 2019
[27]

2025.DOI:10.48550/arXiv.2512.06666

Urav Maniar.The Meta-Learning Gap: Combining Hydra and Quant for Large-Scale Time Series Classification. 2025.DOI:10.48550/arXiv.2512.06666

work page doi:10.48550/arxiv.2512.06666 2025
[28]

HIVE-COTE 2.0: A new meta ensemble for time series classification

Matthew Middlehurst, James Large, Michael Flynn, et al. “HIVE-COTE 2.0: A new meta ensemble for time series classification”. In:Machine Learning110.11–12 (2021), pp. 3211–3243.ISSN: 1573-0565.DOI: 10.1007/ s10994-021-06057-9

2021
[29]

Scikit-learn: Machine Learning in Python

Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, et al. “Scikit-learn: Machine Learning in Python”. In:Journal of Machine Learning Research (JMLR)12.85 (2011), pp. 2825–2830.URL: http://jmlr.org/ papers/v12/pedregosa11a.html

2011
[30]

Temporal Convolutional Neural Network for the Classification of Satellite Image Time Series

Charlotte Pelletier, Geoffrey I. Webb, and François Petitjean. “Temporal Convolutional Neural Network for the Classification of Satellite Image Time Series”. In:Remote Sensing11.5 (2019), p. 523.ISSN: 2072-4292.DOI: 10.3390/rs11050523

work page doi:10.3390/rs11050523 2019
[31]

2022.DOI:10.48550/arXiv.2203.03445

Hojjat Salehinejad, Yang Wang, Yuanhao Yu, et al.S-Rocket: Selective Random Convolution Kernels for Time Series Classification. 2022.DOI:10.48550/arXiv.2203.03445

work page doi:10.48550/arxiv.2203.03445 2022
[32]

Schwartz, J

Roy Schwartz et al. “Green AI”. In:Communications of the ACM63.12 (2020), pp. 54–63.ISSN: 0001-0782. DOI:10.1145/3381831

work page doi:10.1145/3381831 2020
[33]

MultiRocket: multiple pooling operators and transformations for fast and effective time series classification

Chang Wei Tan et al. “MultiRocket: multiple pooling operators and transformations for fast and effective time series classification”. In:Data Mining and Knowledge Discovery36.5 (2022), pp. 1623–1646.ISSN: 1573-756X. DOI:10.1007/s10618-022-00844-1

work page doi:10.1007/s10618-022-00844-1 2022
[34]

Towards Green Automated Machine Learning: Status Quo and Future Directions

Tanja Tornede, Alexander Tornede, Jonas Hanselle, et al. “Towards Green Automated Machine Learning: Status Quo and Future Directions”. In:Journal of Artificial Intelligence Research77 (2023), pp. 427–457.DOI: 10.1613/JAIR.1.14340

work page doi:10.1613/jair.1.14340 2023
[35]

Detach-ROCKET: Sequential Feature Selection for Time Series Classification with Random Convolutional Kernels

Gonzalo Uribarri et al. “Detach-ROCKET: Sequential Feature Selection for Time Series Classification with Random Convolutional Kernels”. In:Data Mining and Knowledge Discovery38.6 (2024), pp. 3922–3947.DOI: 10.1007/s10618-024-01062-7

work page doi:10.1007/s10618-024-01062-7 2024
[36]

Hype, Sustainability, and the Price of the Bigger-is- Better Paradigm in AI

Gael Varoquaux, Sasha Luccioni, and Meredith Whittaker. “Hype, Sustainability, and the Price of the Bigger-is- Better Paradigm in AI”. In:Conference on Fairness, Accountability and Transparency. 2025, pp. 61–75.ISBN: 9798400714825.DOI:10.1145/3715275.3732006

work page doi:10.1145/3715275.3732006 2025
[37]

2017 Time series classification from scratch with deep neural networks: A strong baseline

Zhiguang Wang, Weizhong Yan, and Tim Oates. “Time series classification from scratch with deep neural networks: A strong baseline”. In:International Joint Conference on Neural Networks. 2017, pp. 1578–1585. DOI:10.1109/IJCNN.2017.7966039. (Visited on 02/18/2026)

work page doi:10.1109/ijcnn.2017.7966039 2017
[38]

The Lack of A Priori Distinctions Between Learning Algorithms

David H. Wolpert. “The Lack of A Priori Distinctions Between Learning Algorithms”. In:Neural Computation 8.7 (1996), pp. 1341–1390.DOI:10.1162/neco.1996.8.7.1341. 11 PRUNINGEXTENSIONS ANDEFFICIENCYTRADE-OFFS FORSUSTAINABLETIMESERIESCLASSIFICATION

work page doi:10.1162/neco.1996.8.7.1341 1996
[39]

Sustainable AI: AI for Sustainability and the Sustainability of AI

Aimee van Wynsberghe. “Sustainable AI: AI for Sustainability and the Sustainability of AI”. In:AI and Ethics 1.3 (2021), pp. 213–218.ISSN: 2730-5961.DOI:10.1007/s43681-021-00043-6

work page doi:10.1007/s43681-021-00043-6 2021
[40]

Convolutional Neural Networks for Time Series Classification

Bendong Zhao, Huanzhang Lu, Shangfeng Chen, et al. “Convolutional Neural Networks for Time Series Classification”. In:Journal of Systems Engineering and Electronics28.1 (2017), pp. 162–169.DOI: 10.21629/ JSEE.2017.01.18

2017
[41]

Time Series Classification Using Multi-Channels Deep Convolutional Neural Networks

Yi Zheng, Qi Liu, Enhong Chen, et al. “Time Series Classification Using Multi-Channels Deep Convolutional Neural Networks”. en. In:Web-Age Information Management. 2014, pp. 298–310.ISBN: 978-3-319-08010-9. DOI:10.1007/978-3-319-08010-9_33. 12

work page doi:10.1007/978-3-319-08010-9_33 2014