pith. sign in

arxiv: 1907.09296 · v1 · pith:EFOXF56Gnew · submitted 2019-07-22 · 📡 eess.SP · cs.CV· eess.IV

A-Phase classification using convolutional neural networks

Edgar R. Arce-Santana , Alfonso Alba , Martin O. Mendez , Valdemar Arce-Guevara This is my paper

Pith reviewed 2026-05-24 17:56 UTC · model grok-4.3

classification 📡 eess.SP cs.CVeess.IV
keywords A-phase classificationconvolutional neural networksEEGsleep analysisad-hoc classifiersspectrogramNREM sleep
0
0 comments X

The pith

Training a convolutional neural network on only 25 percent of one subject's A-phases allows it to classify the remaining events at average accuracies of 80.31 percent for detection and over 70 percent for subtypes.

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

The paper shows that instead of building one classifier to work across all people, training a separate convolutional neural network for each subject on only a quarter of their labeled A-phase events produces usable results. Using the log-spectrogram of the EEG as input, these models reach 80.31 percent accuracy distinguishing A-phases from background and 71.87 percent when sorting the three subtypes. Adding a bit more expert-checked data lifts the subtype accuracy to 78.92 percent. This points to a practical way to cut down the expert time needed for sleep EEG annotation by letting the computer handle most events after minimal initial labeling.

Core claim

The authors argue that ad-hoc classifiers based on convolutional neural networks, trained individually for each subject using the log-spectrogram of their EEG signals and only 25 percent of the A-phases, can discriminate A-phases from non-A-phases at 80.31 percent average accuracy and classify A1, A2, A3 subtypes at 71.87 percent, improving to 78.92 percent with additional validated data. This approach is presented as a semi-automatic alternative that requires far less expert effort than full manual review or training a single general model.

What carries the argument

Subject-specific convolutional neural networks that take log-spectrograms of EEG signals as input to classify A-phases.

Load-bearing premise

A-phases exhibit enough consistency in their EEG patterns within each individual that a network trained on a small fraction of them will correctly label the rest of that person's events.

What would settle it

Testing the trained CNN on the held-out 75 percent of A-phases from the same subjects and finding accuracies drop substantially below the reported levels would show the approach does not generalize as claimed.

Figures

Figures reproduced from arXiv: 1907.09296 by Alfonso Alba, Edgar R. Arce-Santana, Martin O. Mendez, Valdemar Arce-Guevara.

Figure 1
Figure 1. Figure 1: Example of A1-, A2- and A3-phases during sleep stage 2 (SS2) and 4 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: EEG Log-spectrogram representation. 3.2.1 Preprocessing In EEG signal analysis, it is often useful to characterize a signal by the time localization of its frequency components. This is particularly important for A￾phase classification, since the A-phase sub-types are defined in terms of their spectral content. A popular time-frequency decomposition technique is the short-time Fourier transform, which has … view at source ↗
Figure 3
Figure 3. Figure 3: Proposed CNN architectures for A-phase classification. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Average accuracy of 20 convolutional networks trained to classify A [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Average accuracy of 20 convolutional networks trained to classify A [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
read the original abstract

A series of short events, called A-phases, can be observed in the human electroencephalogram during NREM sleep. These events can be classified in three groups (A1, A2 and A3) according to their spectral contents, and are thought to play a role in the transitions between the different sleep stages. A-phase detection and classification is usually performed manually by a trained expert, but it is a tedious and time-consuming task. In the past two decades, various researchers have designed algorithms to automatically detect and classify the A-phases with varying degrees of success, but the problem remains open. In this paper, a different approach is proposed: instead of attempting to design a general classifier for all subjects, we propose to train ad-hoc classifiers for each subject using as little data as possible, in order to drastically reduce the amount of time required from the expert. The proposed classifiers are based on deep convolutional neural networks using the log-spectrogram of the EEG signal as input data. Results are encouraging, achieving average accuracies of 80.31% when discriminating between A-phases and non A-phases, and 71.87% when classifying among A-phase sub-types, with only 25% of the total A-phases used for training. When additional expert-validated data is considered, the sub-type classification accuracy increases to 78.92%. These results show that a semi-automatic annotation system with assistance from an expert could provide a better alternative to fully automatic classifiers.

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.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes training subject-specific convolutional neural networks on log-spectrograms of single-channel EEG to classify A-phases (and their A1/A2/A3 subtypes) during NREM sleep. Rather than a single population-level model, the approach trains an ad-hoc CNN per subject on only 25% of that subject's labeled A-phases, reporting average accuracies of 80.31% for binary A-phase vs. non-A-phase discrimination and 71.87% for three-class subtype discrimination across subjects; the subtype accuracy rises to 78.92% when additional expert-validated data are included. The goal is to reduce expert annotation effort while mitigating inter-subject variability.

Significance. If the reported within-subject generalization holds, the work demonstrates a practical route to semi-automatic A-phase annotation that could cut expert review time by roughly 75% while still achieving usable accuracy. The per-subject design is a clear strength relative to prior population-level detectors, and the use of log-spectrogram inputs with CNNs is a standard yet well-motivated choice for this signal-processing task.

major comments (3)
  1. [Methods] Methods section: the procedure for selecting the 25% training events per subject (random sampling, stratification by subtype or sleep stage, or temporal blocking) is not described, nor is any cross-validation scheme or handling of temporal autocorrelation in the EEG; these details are load-bearing for the claim that the held-out accuracies reflect genuine generalization rather than optimistic splits.
  2. [Results] Results section: no baseline classifiers (e.g., SVM on hand-crafted spectral features or existing A-phase detectors from the literature), no statistical tests on the per-subject accuracies, and no confidence intervals or standard deviations across subjects are reported, making it impossible to assess whether 80.31% and 71.87% represent meaningful improvements.
  3. [Results] Results section: class imbalance between A-phases and non-A-phases (and among A1/A2/A3) is not addressed in the training or evaluation protocol; given the typical rarity of A-phases, the reported accuracy figures could be inflated by majority-class performance.
minor comments (2)
  1. [Abstract] Abstract: the phrase 'ad-hoc classifiers' is used without immediate clarification that they are subject-specific; a parenthetical '(i.e., one CNN per subject)' would improve immediate readability.
  2. [Figures] Figure captions (if present): ensure that any spectrogram examples clearly label the frequency and time axes and indicate which events were used for training versus testing.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below, indicating where we agree revisions are warranted and providing clarifications where the manuscript's design choices are defensible.

read point-by-point responses

  1. Referee: [Methods] Methods section: the procedure for selecting the 25% training events per subject (random sampling, stratification by subtype or sleep stage, or temporal blocking) is not described, nor is any cross-validation scheme or handling of temporal autocorrelation in the EEG; these details are load-bearing for the claim that the held-out accuracies reflect genuine generalization rather than optimistic splits.

    Authors: We agree these details should be explicit. The 25% training events were chosen by random sampling independently per subject with no stratification; a single fixed 75/25 split was used to emulate the target semi-automatic workflow rather than cross-validation. Events were treated as independent samples. We will revise the Methods section to document the sampling procedure, the rationale for the single split, and a brief discussion of temporal autocorrelation given the discrete, non-overlapping nature of A-phases. revision: yes

  2. Referee: [Results] Results section: no baseline classifiers (e.g., SVM on hand-crafted spectral features or existing A-phase detectors from the literature), no statistical tests on the per-subject accuracies, and no confidence intervals or standard deviations across subjects are reported, making it impossible to assess whether 80.31% and 71.87% represent meaningful improvements.

    Authors: We will add the standard deviation of per-subject accuracies and simple statistical comparisons against chance-level performance in the revised Results. While the core contribution is the subject-specific limited-data regime rather than head-to-head comparison with population-level detectors, we will include a basic baseline (majority-class and a linear SVM on spectral features) for context. revision: partial

  3. Referee: [Results] Results section: class imbalance between A-phases and non-A-phases (and among A1/A2/A3) is not addressed in the training or evaluation protocol; given the typical rarity of A-phases, the reported accuracy figures could be inflated by majority-class performance.

    Authors: We accept that overall accuracy alone is insufficient given imbalance. The revision will report per-class sensitivity/specificity for the binary task and precision/recall/F1 for the three-class task. No explicit balancing (e.g., oversampling or weighted loss) was applied; this will be stated explicitly along with the new metrics. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper reports empirical results from training per-subject CNN classifiers on 25% of each subject's labeled A-phases and evaluating classification accuracy on the held-out remainder of that subject's events. The central performance numbers are obtained via explicit train-test splits on independent data with no equations, fitted parameters, or self-citations that reduce the reported accuracies to quantities already present in the training inputs by construction. The approach contains no derivation chain, uniqueness theorems, or ansatzes that collapse to the inputs; it is a standard supervised learning evaluation whose claims remain independent of the training data itself.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim depends on standard deep-learning assumptions plus the domain premise that log-spectrograms preserve the spectral distinctions needed for A-phase typing; many architecture and training hyperparameters are chosen to fit the small per-subject datasets.

free parameters (1)
  • CNN architecture and training hyperparameters
    Number of layers, filter sizes, learning rate, and regularization choices are selected to maximize performance on the 25% training split per subject.
axioms (2)
  • domain assumption Log-spectrogram representation of single-channel EEG captures the frequency content distinctions among A1, A2, and A3 phases
    Invoked by the choice of input representation to the CNN.
  • domain assumption A-phases within one subject are sufficiently stationary for a single CNN trained on 25% of events to classify the remainder
    Required for the per-subject training strategy to succeed without additional adaptation.

pith-pipeline@v0.9.0 · 5815 in / 1372 out tokens · 33731 ms · 2026-05-24T17:56:46.205348+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages

  1. [1]

    Sleep and the metabolic syndrome

    Robert Wolk and Virend K Somers. Sleep and the metabolic syndrome. Experimental Physiology, 92(1):67–78, 2007

    work page 2007

  2. [2]

    Cumula- tive sleepiness, mood disturbance, and psychomotor vigilance performance decrements during a week of sleep restricted to 4–5 hours per night

    David F Dinges, Frances Pack, Katherine Williams, Kelly A Gillen, John W Powell, Geoffrey E Ott, Caitlin Aptowicz, and Allan I Pack. Cumula- tive sleepiness, mood disturbance, and psychomotor vigilance performance decrements during a week of sleep restricted to 4–5 hours per night. Sleep, 20(4):267–277, 1997

    work page 1997

  3. [3]

    Sleep disorders and sleep depri- vation: an unmet public health problem

    Bruce M Altevogt, Harvey R Colten, et al. Sleep disorders and sleep depri- vation: an unmet public health problem . National Academies Press, 2006

    work page 2006

  4. [4]

    The aasm manual for the scoring of sleep and associated events: Rules

    Conrad Iber. The aasm manual for the scoring of sleep and associated events: Rules. Terminology and Technical Specification, 2007

    work page 2007

  5. [5]

    Atlas, rules, and recording techniques for the scoring of cyclic alternating pattern (cap) in human sleep

    Mario Giovanni Terzano, Liborio Parrino, Arianna Smerieri, Ronald Chervin, Sudhansu Chokroverty, Christian Guilleminault, Max Hir- shkowitz, Mark Mahowald, Harvey Moldofsky, Agostino Rosa, et al. Atlas, rules, and recording techniques for the scoring of cyclic alternating pattern (cap) in human sleep. Sleep medicine, 3(2):187–199, 2002

    work page 2002

  6. [6]

    Clinical applications of cyclic alternating pattern

    Mario Giovanni Terzano and Liborio Parrino. Clinical applications of cyclic alternating pattern. Physiology & behavior, 54(4):807–813, 1993. 15

    work page 1993

  7. [7]

    Inter-rater reliability of sleep cyclic al- ternating pattern (cap) scoring and validation of a new computer-assisted cap scoring method

    Raffaele Ferri, Oliviero Bruni, Silvia Miano, Arianna Smerieri, Karen Spruyt, and Mario G Terzano. Inter-rater reliability of sleep cyclic al- ternating pattern (cap) scoring and validation of a new computer-assisted cap scoring method. Clinical neurophysiology, 116(3):696–707, 2005

    work page 2005

  8. [8]

    All-night eeg power spectral analysis of the cyclic alternating pattern components in young adult subjects

    Raffaele Ferri, Oliviero Bruni, Silvia Miano, Giuseppe Plazzi, and Mario G Terzano. All-night eeg power spectral analysis of the cyclic alternating pattern components in young adult subjects. Clinical Neurophysiology, 116(10):2429–2440, 2005

    work page 2005

  9. [9]

    Quantitative analysis of sleep eeg microstructure in the time– frequency domain

    Fabrizio De Carli, Lino Nobili, Manolo Beelke, Tsuyoshi Watanabe, Ar- ianna Smerieri, Liborio Parrino, Mario Giovanni Terzano, and Franco Ferrillo. Quantitative analysis of sleep eeg microstructure in the time– frequency domain. Brain Research Bulletin, 63(5):399–405, 2004

    work page 2004

  10. [10]

    An automatic method for the recognition and classification of the a-phases of the cyclic alternating pat- tern

    Carlo Navona, Umberto Barcaro, Enrica Bonanni, Fabio Di Martino, Michelangelo Maestri, and Luigi Murri. An automatic method for the recognition and classification of the a-phases of the cyclic alternating pat- tern. Clinical neurophysiology, 113(11):1826–1831, 2002

    work page 2002

  11. [11]

    Automatic detection of cap on central and fronto-central eeg leads via support vector machines

    Sara Mariani, Andrea Grassi, Martin O Mendez, Liborio Parrino, Mario G Terzano, and Anna M Bianchi. Automatic detection of cap on central and fronto-central eeg leads via support vector machines. In 2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pages 1491–1494. IEEE, 2011

    work page 2011

  12. [12]

    Efficient automatic classifiers for the de- tection of a phases of the cyclic alternating pattern in sleep

    Sara Mariani, Elena Manfredini, Valentina Rosso, Andrea Grassi, Mar- tin O Mendez, Alfonso Alba, Matteo Matteucci, Liborio Parrino, Mario G Terzano, Sergio Cerutti, et al. Efficient automatic classifiers for the de- tection of a phases of the cyclic alternating pattern in sleep. Medical & biological engineering & computing, 50(4):359–372, 2012

    work page 2012

  13. [13]

    Deep learning

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep learning. MIT press, 2016

    work page 2016

  14. [14]

    Deep learning for healthcare applications based on physiological signals: A review

    Oliver Faust, Yuki Hagiwara, Tan Jen Hong, Oh Shu Lih, and U Rajendra Acharya. Deep learning for healthcare applications based on physiological signals: A review. Computer methods and programs in biomedicine, 161:1– 13, 2018

    work page 2018

  15. [15]

    A survey on deep learning in medical image analysis

    Geert Litjens, Thijs Kooi, Babak Ehteshami Bejnordi, Arnaud Arindra Adiyoso Setio, Francesco Ciompi, Mohsen Ghafoorian, Jeroen Awm Van Der Laak, Bram Van Ginneken, and Clara I S´ anchez. A survey on deep learning in medical image analysis. Medical image analysis, 42:60–88, 2017

    work page 2017

  16. [16]

    Deep learning in medical image analysis

    Dinggang Shen, Guorong Wu, and Heung-Il Suk. Deep learning in medical image analysis. Annual review of biomedical engineering, 19:221–248, 2017. 16

    work page 2017

  17. [17]

    A novel wavelet sequence based on deep bidirectional lstm network model for ecg signal classification

    ¨Ozal Yildirim. A novel wavelet sequence based on deep bidirectional lstm network model for ecg signal classification. Computers in biology and medicine, 96:189–202, 2018

    work page 2018

  18. [18]

    A deep convolutional neural network model to classify heartbeats

    U Rajendra Acharya, Shu Lih Oh, Yuki Hagiwara, Jen Hong Tan, Muham- mad Adam, Arkadiusz Gertych, and Ru San Tan. A deep convolutional neural network model to classify heartbeats. Computers in biology and medicine, 89:389–396, 2017

    work page 2017

  19. [19]

    Ar- rhythmia detection using deep convolutional neural network with long du- ration ecg signals

    ¨Ozal Yıldırım, Pawe l P lawiak, Ru-San Tan, and U Rajendra Acharya. Ar- rhythmia detection using deep convolutional neural network with long du- ration ecg signals. Computers in biology and medicine , 102:411–420, 2018

    work page 2018

  20. [20]

    A deep learn- ing approach for parkinson’s disease diagnosis from eeg signals

    Shu Lih Oh, Yuki Hagiwara, U Raghavendra, Rajamanickam Yuvaraj, N Arunkumar, M Murugappan, and U Rajendra Acharya. A deep learn- ing approach for parkinson’s disease diagnosis from eeg signals. Neural Computing and Applications , pages 1–7, 2018

    work page 2018

  21. [21]

    Deep convolutional neural network for the automated detec- tion and diagnosis of seizure using eeg signals

    U Rajendra Acharya, Shu Lih Oh, Yuki Hagiwara, Jen Hong Tan, and Hojjat Adeli. Deep convolutional neural network for the automated detec- tion and diagnosis of seizure using eeg signals. Computers in biology and medicine, 100:270–278, 2018

    work page 2018

  22. [22]

    A deep con- volutional neural network model for automated identification of abnormal eeg signals

    ¨Ozal Yıldırım, Ulas Baran Baloglu, and U Rajendra Acharya. A deep con- volutional neural network model for automated identification of abnormal eeg signals. Neural Computing and Applications , pages 1–12, 2018

    work page 2018

  23. [23]

    Deep neural architectures for mapping scalp to intracranial eeg

    Andreas Antoniades, Loukianos Spyrou, David Martin-Lopez, Antonio Valentin, Gonzalo Alarcon, Saeid Sanei, and Clive Cheong Took. Deep neural architectures for mapping scalp to intracranial eeg. International journal of neural systems , 28(08):1850009, 2018

    work page 2018

  24. [24]

    Deepsleepnet: A model for automatic sleep stage scoring based on raw single-channel eeg

    Akara Supratak, Hao Dong, Chao Wu, and Yike Guo. Deepsleepnet: A model for automatic sleep stage scoring based on raw single-channel eeg. IEEE Transactions on Neural Systems and Rehabilitation Engineer- ing, 25(11):1998–2008, 2017

    work page 1998

  25. [25]

    Automatic sleep stage scoring using time-frequency analysis and stacked sparse autoencoders

    Orestis Tsinalis, Paul M Matthews, and Yike Guo. Automatic sleep stage scoring using time-frequency analysis and stacked sparse autoencoders. An- nals of biomedical engineering , 44(5):1587–1597, 2016

    work page 2016

  26. [26]

    Use of features from rr-time series and eeg signals for automated classification of sleep stages in deep neural network framework

    RK Tripathy and U Rajendra Acharya. Use of features from rr-time series and eeg signals for automated classification of sleep stages in deep neural network framework. Biocybernetics and Biomedical Engineering, 38(4):890– 902, 2018

    work page 2018

  27. [27]

    A deep learning architecture for temporal sleep stage classification using multivariate and multimodal time series

    Stanislas Chambon, Mathieu N Galtier, Pierrick J Arnal, Gilles Wainrib, and Alexandre Gramfort. A deep learning architecture for temporal sleep stage classification using multivariate and multimodal time series. IEEE 17 Transactions on Neural Systems and Rehabilitation Engineering, 26(4):758– 769, 2018

    work page 2018

  28. [28]

    Cascaded lstm recurrent neural network for automated sleep stage classification using single-channel eeg signals

    Nicola Michielli, U Rajendra Acharya, and Filippo Molinari. Cascaded lstm recurrent neural network for automated sleep stage classification using single-channel eeg signals. Computers in biology and medicine , 106:71–81, 2019

    work page 2019

  29. [29]

    Automatic detection of cyclic alternating pattern (cap) sequences in sleep: preliminary results

    AC Rosa, L Parrino, and MG Terzano. Automatic detection of cyclic alternating pattern (cap) sequences in sleep: preliminary results. Clinical neurophysiology, 110(4):585–592, 1999

    work page 1999

  30. [30]

    A model-based detector of vertex waves and k complexes in sleep elec- troencephalogram

    AC Rosa, B Kemp, T Paiva, FH Lopes da Silva, and HAC Kamphuisen. A model-based detector of vertex waves and k complexes in sleep elec- troencephalogram. Electroencephalography and clinical neurophysiology , 78(1):71–79, 1991

    work page 1991

  31. [31]

    A general automatic method for the analysis of nrem sleep microstructure

    Umberto Barcaro, Enrica Bonanni, Michelangelo Maestri, Luigi Murri, Li- borio Parrino, and Mario Giovanni Terzano. A general automatic method for the analysis of nrem sleep microstructure. Sleep Medicine, 5(6):567–576, 2004

    work page 2004

  32. [32]

    On separability of a-phases during the cyclic alternating pattern

    Martin O Mendez, Alfonso Alba, Ioanna Chouvarda, Guilia Milioli, Andrea Grassi, Mario G Terzano, and Liborio Parrino. On separability of a-phases during the cyclic alternating pattern. In 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pages 2253–2256. IEEE, 2014

    work page 2014

  33. [33]

    Analysis of a-phase transitions during the cyclic alternating pattern under normal sleep

    Martin Oswaldo Mendez, Ioanna Chouvarda, Alfonso Alba, Anna Maria Bianchi, Andrea Grassi, Edgar Arce-Santana, Guilia Milioli, Mario Gio- vanni Terzano, and Liborio Parrino. Analysis of a-phase transitions during the cyclic alternating pattern under normal sleep. Medical & biological engineering & computing, 54(1):133–148, 2016

    work page 2016

  34. [34]

    Presenting efficient features for automatic cap detection in sleep eeg signals

    Foroozan Karimzadeh, Esmaeil Seraj, Reza Boostani, and Mohammad Torabi-Nami. Presenting efficient features for automatic cap detection in sleep eeg signals. In 2015 38th International Conference on Telecommuni- cations and Signal Processing (TSP) , pages 448–452. IEEE, 2015

    work page 2015

  35. [35]

    Automatic detection of cyclic alter- nating pattern

    F´ abio Mendon¸ ca, Ana Fred, Sheikh Shanawaz Mostafa, Fernando Morgado- Dias, and Antonio G Ravelo-Garc´ ıa. Automatic detection of cyclic alter- nating pattern. Neural Computing and Applications , pages 1–11, 2018

    work page 2018

  36. [36]

    A-phases subtype detection using different classification methods

    F´ atima Machado, C´ esar Teixeira, Clara Santos, Concei¸ c˜ ao Bento, Francisco Sales, and Antonio Dourado. A-phases subtype detection using different classification methods. In 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) , pages 1026–1029. IEEE, 2016. 18

    work page 2016

  37. [37]

    A knowledge discovery methodology from eeg data for cyclic alternating pattern detection

    F´ atima Machado, Francisco Sales, Clara Santos, Ant´ onio Dourado, and CA Teixeira. A knowledge discovery methodology from eeg data for cyclic alternating pattern detection. Biomedical engineering online , 17(1):185, 2018

    work page 2018

  38. [38]

    Combination of deep and shallow networks for cyclic alternating patterns detection

    Sheikh Shanawaz Mostafa, F´ abio Mendon¸ ca, Antonio Ravelo-Garc´ ıa, and Fernando Morgado-Dias. Combination of deep and shallow networks for cyclic alternating patterns detection. In 2018 13th APCA International Conference on Control and Soft Computing (CONTROLO) , pages 98–103. IEEE, 2018

    work page 2018

  39. [39]

    Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals

    Ary L Goldberger, Luis AN Amaral, Leon Glass, Jeffrey M Hausdorff, Plamen Ch Ivanov, Roger G Mark, Joseph E Mietus, George B Moody, Chung-Kang Peng, and H Eugene Stanley. Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals. Circulation, 101(23):e215–e220, 2000

    work page 2000

  40. [40]

    Haghighi-Mood and J.N

    A. Haghighi-Mood and J.N. Torry. Time frequency analysis of systolic murmurs. time-frequency analysis of biomedical signals. In IEE Colloquium on Year, pages 2/1–2/3. IEE, 1997

    work page 1997

  41. [41]

    Krizhevsky, I Sutskever, and G

    A. Krizhevsky, I Sutskever, and G. E. Hinton. Imagenet classification with deep convolutional neural networks. In NIPS’12 Proceedings of the 25th International Conference on Neural Information Processing Systems, pages 1097–1105. NIPS, 2012

    work page 2012

  42. [42]

    I. Arel, D. Rose, and T. Karnowski. Deep machine learninga new frontier in artificial intelligence research [research frontier]. IEEE Comput. Int. Mag. , 5:13–18, 2010

    work page 2010

  43. [43]

    Ciresan, U

    D. Ciresan, U. Meier, and J. Schmidhuber. Multi-column deep neural net- works for image classification. In 2012 Computer Vision and Pattern Recog- nition (CVPR), pages 3642–3649. IEEE, 2012. 19

    work page 2012