跳到主要內容

臺灣博碩士論文加值系統

(44.192.94.177) 您好!臺灣時間:2024/07/21 20:17
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

我願授權國圖
: 
twitterline
研究生:段惟芳
研究生(外文):Duy-Phuong Doan
論文名稱:利用機器學習和深度學習技術對神經傳導數據進行分類
論文名稱(外文):Classification of conduction studies of median nerve using machine learning and deep learning techniques
指導教授:黃彥華黃彥華引用關係
指導教授(外文):Yen-Hua Huang
學位類別:碩士
校院名稱:國立陽明大學
系所名稱:生物醫學資訊研究所
學門:生命科學學門
學類:生物化學學類
論文種類:學術論文
論文出版年:2020
畢業學年度:108
語文別:英文
論文頁數:107
中文關鍵詞:神經傳導研究機器學習深度學習生成對抗網路
外文關鍵詞:Nerve conduction studiesMachine learningDeep learningGenerative adversarial networks
相關次數:
  • 被引用被引用:0
  • 點閱點閱:171
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
Table of Contents
Declaration of Authorship i
Mandarin abstract ii
English abstract iii
Acknowledgments v
Table of Contents vi
List of Figures ix
List of Tables xiii
List of Abbreviations xiv
Chapter 1. Introduction 1
1.1. Electrodiagnostic studies (EDX) 1
Nerve conduction studies (NCS) 2
1.2. Median nerve 3
1.3. Aims 4
Chapter 2. Literature review 5
2.1. Computational approaches on time-series medical data 5
2.2. Convolutional neural networks on medical data 5
2.2.1. Convolutional neural networks (CNN) 5
2.2.2. Application of CNN on biomedical data 6
2.3. Transfer learning on building deep learning model 6
2.4. Autoencoder for learning feature representation in medical data 7
2.5. Generative adversarial networks 8
2.5.1. Generative adversarial networks for data augmentation for medical data 10
2.5.1.1. Deep convolutional GAN 10
2.5.1.2. Wasserstein GAN 11
2.5.1.3. Wasserstein with gradient penalty GAN 11
2.5.2. Generative adversarial networks for unsupervised learning 11
2.5.2.1. Sparsity constrained generative adversarial networks (Sparse-GAN) 12
Chapter 3. Materials and Methods 17
3.1. Materials 18
3.2. Conversion of the diagnosis into structural labeling 19
3.2.1. Workflow of building the encryption model 19
3.2.2. Natural language processing (NLP) 20
3.2.3. Machine learning application on building encryption models 20
3.2.3.1. Support vector machine (SVM) 20
3.2.3.2. Random forest classifier (RF) 20
3.2.3.3. Multinomial Naïve Bayes 21
3.2.3.4. Logistic regression 21
3.2.4. Deep learning application on building encryption models 21
3.2.4.1. Recurrent neural networks (RNN) 22
3.2.4.2. Long-short term memory networks (LSTM) 22
3.3. Building a classification model on the NCS numeric data 23
3.3.1. Choosing targets for the classification model 24
3.3.1.1. Structured missing data 24
3.3.1.2. Select the suitable target for the classification model 25
3.3.2. Data preprocessing for classification model construction 25
3.3.3. Building classification models for classify median numeric data 27
3.3.3.1. T-distributed Stochastic Neighbor Embedding (t-SNE) for visualization 27
3.3.3.2. Support Vector Machine (SVM) 28
3.3.3.3. Random Forests (RF) 28
3.3.3.4. eXtreme Gradient Boosting (XGBoost) 28
3.4. Building a classification model on the NCS curve data 29
3.4.1. Data collection 30
3.4.2. Data preprocessing for building a classification model 30
3.4.3. Building classification model for the curve data of the median nerve 32
3.4.4. Convolutional Neural Networks (CNN) 33
3.4.5. Autoencoder (AE) 33
3.4.6. Generative Adversarial Networks (GAN) 34
3.4.6.1. Deep convolutional generative adversarial networks (DCGAN) 34
3.4.6.2. Wasserstein generative adversarial networks (WGAN) 35
3.4.6.3. Wasserstein generative adversarial networks with gradient penalty (WGAN-GP) 36
3.4.6.4. Unsupervised anomaly detection using Sparsity-constrained generative adversarial networks (Sparse-GAN) 37
Chapter 4. Results and Discussion 38
4.1. Conversion of the diagnosis into structural labeling using natural language processing (NLP) technologies 38
4.1.1. Data preprocessing for the encryption model 39
4.1.1.1. Remove stop words and punctuations 39
4.1.1.2. Stemming and lemmatization 40
4.1.1.3. Creating a vector space model 41
4.1.1.4. Data imbalance in the data set 41
4.1.2. Machine learning algorithms selection and application 43
4.1.2.1. Performance of the encryption model on the Code 1 group 44
4.1.2.1. Performance of the encryption model on the Code 2 group 46
4.1.2.1. Performance of the encryption model on the Code 3 group 48
4.1.2.1. Performance of the encryption model on the Code 4 group 49
4.1.3. Deep learning model application 52
4.2. Creating diagnosis models based on NCS numeric data 54
4.2.1. Data visualization using t-SNE 54
4.2.2. Support Vectors Machine (SVM) Results 57
4.2.3. Random Forests (RF) Results 58
4.2.4. XGBoost Results 60
4.2.5. Features selection and hyperparameter tuning 62
4.2.5.1. Hyperparameter tuning 63
4.2.5.2. Drop the most crucial feature 64
4.2.5.3. Drop the second most important feature 65
4.2.5.4. Filter out highly correlated features 67
4.3. Creating diagnosis models based on NCS curve data 71
4.3.1. Data visualization using Uniform Manifold Approximation and Projection (UMAP) 71
4.3.2. Baseline model using traditional machine learning techniques 72
4.3.3. Classification model using Convolutional Neural Networks on sensory nerve data 73
4.3.4. Applying transfer learning technique 75
4.3.4.1. Autoencoder (AE) training 75
4.3.4.2. Transfer learning model using pre-trained AE 79
4.3.5. Applying generative adversarial networks (GAN) 81
4.3.5.1. Generate normal curve data using GAN 81
4.3.5.2. Unsupervised anomaly detection using sparse-GAN 92
Chapter 5. Conclusion and Future work 101
5.1. Conclusion 101
5.2. Future work 102
References 103
Appendix I

List of Figures
Figure 1. Anatomy of the peripheral nervous system [2]. 1
Figure 2. Different types of NCS measurements [2]. 2
Figure 3. Anatomy of median nerve [5] 3
Figure 4. The illustration of a sample CNN with 2 convolution and one fully-connected layers [9] 6
Figure 5. The curves of the loss function. A. The curve of function (2). B. The curve of function (3) 9
Figure 6. Structure of the sparse-GAN model [28] 13
Figure 7. PatchGAN discriminator [30]. 14
Figure 8. Overall workflow of the study 17
Figure 9. Examples format of the NCS numeric data 19
Figure 10. Workflow of building an encryption model 19
Figure 11. An example structure of the RNN model 22
Figure 12. Structure of the RNN model 22
Figure 13. Structure of the LSTM unit [43]. 23
Figure 14. Examples of NCS numeric values calculated from NCS curve data [2] 24
Figure 15. Missing value percentages in the NCS numeric data 25
Figure 16. Workflow for the preprocessing of the NCS numeric data. The number of features means the number of curve features being retained. 26
Figure 17. Workflow for building the classification model. 27
Figure 18. Preprocessing process for the NCS curve data. 31
Figure 19. An example of the NCS curve data 31
Figure 20. Workflow for building a classification model on the NCS curve data. 32
Figure 21. Classification model using 1D CNN 33
Figure 22. A sample structure of the GAN model 34
Figure 23. The structure of the DCGAN model 35
Figure 24. The structure of the WGAN model 36
Figure 25. Structure of the WGAN-GP model 37
Figure 26. The class distribution of the first NCS numeric dataset. 42
Figure 27. The accuracy, precision, and recall of different machine learning models trained for 100 times. Each point represents a single model. 44
Figure 28. Confusion matrix of the first model, A. Using Count Vectorizer, B. Using Tfidf Vectorizer. 46
Figure 29. Confusion matrix of the second model, A. Using Count Vectorizer, B. Using Tfidf Vectorizer. 47
Figure 30. Confusion matrix of the third model, A. Using Count Vectorizer, B. Using Tfidf Vectorizer. 49
Figure 31. Confusion matrix of the fourth model, A. Using Count Vectorizer, B. Using Tfidf Vectorizer. 50
Figure 32. Recurrent neural network (left) and Long-short term memory (right) accuracy and loss on the train and validation data sets. 52
Figure 33. Confusion matrix of the LSTM model 53
Figure 34. Data visualization and distribution using t-SNE. A. Before removing outliers. B. After removing outliers. Colored in red is the normal data point (labeled as 0), and in blue is the abnormal (labeled as 1). 54
Figure 35. Testing t-SNE with different perplexity values. A. The perplexity of 30. B. The perplexity of 40. C. The perplexity of 50. 56
Figure 36. Classification model performance using Support Vector Machine (SVM) with raw input number and input number with scaling. Performance measured using Matthew Correlation Coefficient (MCC) and Accuracy (acc). Each point represents one single trained model. 57
Figure 37. Normalized Feature Importance of different features calculated by the Random Forests model 59
Figure 38. Classification model performance using Random Forests (RF) with raw input number and input number with scaling. Performance measured using Matthew Correlation Coefficient (MCC) and Accuracy (acc). Each point represents one single trained model. 60
Figure 39. Classification model performance using different machine learning methods and XGBoost. MCC score and Accuracy were used as the evaluation metrics. Each point represents one single trained model. 61
Figure 40. XGBoost model performance on our dataset. A. Normalized Feature Importance of the XGBoost model. B. The confusion matrix of one XGBoost model. The y-axis represents the ground truth, while the x-axis represents the predicted labels. 0 stands for normal, and 1 stands for abnormal. 62
Figure 41. The performance of the XGBoost model before and after performing hyperparameter tuning. Accuracy and MCC score were used as the evaluation metrics. Each point represents a single trained model 63
Figure 42. XGBoost model performance after dropping the most important feature. A. Normalized Feature Importance of the XGBoost model. B. The confusion matrix of one XGBoost model. The y-axis represents the ground truth, while the x-axis represents the predicted labels. 0 stands for normal, and 1 stands for abnormal. 65
Figure 43. XGBoost model performance after dropping the second most important feature. A. Normalized Feature Importance of the XGBoost model. B. The confusion matrix of one XGBoost model. The y-axis represents the ground truth, while the x-axis represents the predicted labels. 0 stands for normal, and 1 stands for abnormal. 66
Figure 44. The cluster map of the Spearman correlation between different features of the NCS numeric data. The x-down and y-right contain all the features' names, x-up and y-left contain the hierarchical clustering information. The color map represents the correlation degree. 67
Figure 45. XGBoost model performance after dropping all the distal related parameters. A. Normalized Feature Importance of the XGBoost model. B. The confusion matrix of one XGBoost model. The y-axis represents the ground truth, while the x-axis represents the predicted labels. 0 stands for normal, and 1 stands for abnormal. 68
Figure 46. XGBoost model performance after dropping all the proximal related parameters. A. Normalized Feature Importance of the XGBoost model. B. The confusion matrix of one XGBoost model. The y-axis represents the ground truth, while the x-axis represents the predicted labels. 0 stands for normal, and 1 stands for abnormal. 70
Figure 47. XGBoost model performance using different feature subsets. MCC score and Accuracy were used as the evaluation metrics. Each point represents one single trained model. XGB: original model, XGB_d_first: drop the most important feature, XGB_d_second: drop the second most important feature, XGB_d_distal: drop all the distal features, XGB_d_proximal: drop all the proximal features. 70
Figure 48. Two-dimensional representation of the sensory nerve curve data using UMAP 72
Figure 49. Accuracy and Matthew correlation coefficient score of the traditional machine learning techniques. 73
Figure 50. CNN classification model training accuracy and loss 74
Figure 51. CNN classification model performance for 100 times 74
Figure 52. Dense autoencoder performance. A. Train loss, B. Test loss, C. Examples of the original and reconstructed images 76
Figure 53. Long-short term memory autoencoder performance. A. Train loss, B. Test loss, C. Examples of the original and reconstructed images 77
Figure 54. Convolutional autoencoder performance. A. Train loss, B. Test loss, C. Examples of the original, code layer, and reconstructed images 78
Figure 55. Loss curves of the transferred learning model 80
Figure 56. The classification model performance with and without transfer learning 81
Figure 57. Images generated by DCGAN model 82
Figure 58. Images generated by WGAN model with clipping value of 0.1 84
Figure 59. Loss of the WGAN model with clipping value equals 0.01 85
Figure 60. Images generated by WGAN models with clipping value of 10 86
Figure 61. Loss of the WGAN model with clipping value equals 10 87
Figure 62. Images generated by WGAN-GP model 88
Figure 63. Loss of the WGAN-GP model 89
Figure 64. Images generated by WGAN-GP-LSTM model 90
Figure 65. Loss of the WGAN-GP-LSTM model 91
Figure 66. Loss of the Sparse-GAN model with original structure 93
Figure 67. Performance of the Sparse-GAN model with original structure on validation dataset 94
Figure 68. Different ranges of the curve signals 96
Figure 69. Loss of the Sparse-GAN model with simplified structure on preprocessed and normalized data 97
Figure 70. Performance of the Sparse-GAN model with simplified structure on preprocessed and normalized data 98
Figure 71. Loss of the Sparse-GAN model with simplified structure on shorten curve data 99
Figure 72. Performance of the Sparse-GAN model with simplified structure on shorten curve data 100
Figure 73. Data distribution of the normal data on five difference age ranges of the male and female groups. IV

List of Tables
Table 1. Data statistics for the NCS curve of the median nerve 18
Table 2. List of the Median nerve features selected for downstream processes. 26
Table 3. Example for changing the encoding to create four different labels. 39
Table 4. Filtered out diagnosis or coding. 39
Table 5. Abbreviated words list. 40
Table 6. Overall performance of four different models combined 51
Table 7. An example of the predictions combined from four models 51
Table 8. Different SVM model performance with the number of iterations, mean value, standard deviation, and max value. 57
Table 9. Different RF model performance with the number of iterations, mean value, standard deviation, and max value. 60
Table 10. Tuning the dense layer in the transfer learning model 79
Table 11. Time taken for training GAN models 91
Table 12. Performance of different approach for building the Sparse-GAN model 95
Table 13. Statistical analysis on the normal NCS data II
Table 14. Correlation between Median features and demographic data. IV
References
[1] M. J. Kothari, M. A. Blakeslee, R. Reichwein, Z. Simmons, and E. L. Logigian, “Electrodiagnostic studies: Are they useful in clinical practice?,” Arch. Phys. Med. Rehabil., vol. 79, no. 12, pp. 1510–1511, Dec. 1998, doi: 10.1016/S0003-9993(98)90411-7.
[2] “Electrodiagnostic Studies | Clinical Gate.” https://clinicalgate.com/electrodiagnostic-studies/ (accessed Jul. 16, 2020).
[3] R. E. Ward et al., “Sensory and motor peripheral nerve function and incident mobility disability,” J. Am. Geriatr. Soc., vol. 62, no. 12, pp. 2273–2279, Dec. 2014, doi: 10.1111/jgs.13152.
[4] M. Soubeyrand, R. Melhem, M. Protais, M. Artuso, and M. Crézé, “Anatomy of the median nerve and its clinical applications,” Hand Surgery and Rehabilitation, vol. 39, no. 1. Elsevier Masson SAS, pp. 2–18, Feb. 01, 2020, doi: 10.1016/j.hansur.2019.10.197.
[5] C. Bruno and S. McCaughan, “Median Nerve Block,” in Pain, Springer International Publishing, 2019, pp. 477–480.
[6] F. Yaghouby, A. Ayatollahi, R. Bahramali, M. Yaghouby, and A. H. Alavi, “Towards automatic detection of atrial fibrillation: A hybrid computational approach,” Comput. Biol. Med., vol. 40, no. 11–12, pp. 919–930, Nov. 2010, doi: 10.1016/j.compbiomed.2010.10.004.
[7] Q. Li, C. Rajagopalan, and G. D. Clifford, “A machine learning approach to multi-level ECG signal quality classification,” Comput. Methods Programs Biomed., vol. 117, no. 3, pp. 435–447, Dec. 2014, doi: 10.1016/j.cmpb.2014.09.002.
[8] Y. Lecun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015, doi: 10.1038/nature14539.
[9] S. Kiranyaz, O. Avci, O. Abdeljaber, T. Ince, M. Gabbouj, and D. J. Inman, “1D Convolutional Neural Networks and Applications: A Survey,” pp. 1–20, 2019, [Online]. Available: http://arxiv.org/abs/1905.03554.
[10] B. B. Barıs¸kayalıbay, G. Jensen, and P. Van Der Smagt, “CNN-based Segmentation of Medical Imaging Data.” [Online]. Available: https://github.com/BRML/.
[11] M. Sajjad, S. Khan, K. Muhammad, W. Wu, A. Ullah, and S. W. Baik, “Multi-grade brain tumor classification using deep CNN with extensive data augmentation,” J. Comput. Sci., vol. 30, pp. 174–182, Jan. 2019, doi: 10.1016/j.jocs.2018.12.003.
[12] Y. Yu, H. Lin, J. Meng, X. Wei, H. Guo, and Z. Zhao, “Deep Transfer Learning for Modality Classification of Medical Images,” Information, vol. 8, no. 3, p. 91, Jul. 2017, doi: 10.3390/info8030091.
[13] H. C. Shin et al., “Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning,” IEEE Trans. Med. Imaging, vol. 35, no. 5, pp. 1285–1298, May 2016, doi: 10.1109/TMI.2016.2528162.
[14] P. Vincent, H. Larochelle, Y. Bengio, and P.-A. Manzagol, “Extracting and Composing Robust Features with Denoising Autoencoders.”
[15] Y. Bengio, A. Courville, and P. Vincent, “Representation Learning: A Review and New Perspectives.” [Online]. Available: http://www.image-net.org/challenges/LSVRC/2012/results.html.
[16] A. M. Karim, M. S. Güzel, M. R. Tolun, H. Kaya, and F. V. Çelebi, “A new framework using deep auto-encoder and energy spectral density for medical waveform data classification and processing,” Biocybern. Biomed. Eng., vol. 39, no. 1, pp. 148–159, Jan. 2019, doi: 10.1016/j.bbe.2018.11.004.
[17] H. C. Shin, M. Orton, D. J. Collins, S. Doran, and M. O. Leach, “Autoencoder in time-series analysis for unsupervised tissues characterisation in a large unlabelled medical image dataset,” in Proceedings - 10th International Conference on Machine Learning and Applications, ICMLA 2011, Dec. 2011, vol. 1, pp. 259–264, doi: 10.1109/ICMLA.2011.38.
[18] I. Goodfellow et al., “Generative Adversarial Nets,” in Advances in Neural Information Processing Systems 27, Z. Ghahramani, M. Welling, C. Cortes, N. D. Lawrence, and K. Q. Weinberger, Eds. Curran Associates, Inc., 2014, pp. 2672–2680.
[19] X. Huang, Y. Li, O. Poursaeed, J. Hopcroft, and S. Belongie, “Stacked Generative Adversarial Networks,” Proc. - 30th IEEE Conf. Comput. Vis. Pattern Recognition, CVPR 2017, vol. 2017-January, pp. 1866–1875, Dec. 2016, Accessed: Jul. 18, 2020. [Online]. Available: http://arxiv.org/abs/1612.04357.
[20] A. Shrivastava, T. Pfister, O. Tuzel, J. Susskind, W. Wang, and R. Webb, “Learning from Simulated and Unsupervised Images through Adversarial Training,” Proc. - 30th IEEE Conf. Comput. Vis. Pattern Recognition, CVPR 2017, vol. 2017-January, pp. 2242–2251, Dec. 2016, Accessed: Jul. 18, 2020. [Online]. Available: http://arxiv.org/abs/1612.07828.
[21] A. Radford, L. Metz, and S. Chintala, “Unsupervised representation learning with deep convolutional generative adversarial networks,” 4th Int. Conf. Learn. Represent. ICLR 2016 - Conf. Track Proc., pp. 1–16, 2016.
[22] M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein GAN,” 2017, [Online]. Available: http://arxiv.org/abs/1701.07875.
[23] D. Nie et al., “Medical Image Synthesis with Deep Convolutional Adversarial Networks,” IEEE Trans. Biomed. Eng., vol. 65, no. 12, pp. 2720–2730, Dec. 2018, doi: 10.1109/TBME.2018.2814538.
[24] H.-C. Shin et al., “Medical Image Synthesis for Data Augmentation and Anonymization using Generative Adversarial Networks.” [Online]. Available: www.adni-info.org.
[25] I. Gulrajani, F. Ahmed, M. Arjovsky, V. Dumoulin, and A. Courville, “Improved training of wasserstein GANs,” Adv. Neural Inf. Process. Syst., vol. 2017-Decem, pp. 5768–5778, 2017.
[26] B. Hu, Y. Tang, E. I.-C. Chang, Y. Fan, M. Lai, and Y. Xu, “Unsupervised Learning for Cell-level Visual Representation in Histopathology Images with Generative Adversarial Networks,” IEEE J. Biomed. Heal. Informatics, vol. 23, no. 3, pp. 1316–1328, Nov. 2017, doi: 10.1109/JBHI.2018.2852639.
[27] T. Schlegl, P. Seeböck, S. M. Waldstein, U. Schmidt-Erfurth, and G. Langs, “Unsupervised anomaly detection with generative adversarial networks to guide marker discovery,” in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2017, vol. 10265 LNCS, pp. 146–147, doi: 10.1007/978-3-319-59050-9_12.
[28] K. Zhou et al., “Sparse-Gan: Sparsity-Constrained Generative Adversarial Network for Anomaly Detection in Retinal OCT Image,” Proc. - Int. Symp. Biomed. Imaging, vol. 2020-April, pp. 1227–1231, 2020, doi: 10.1109/ISBI45749.2020.9098374.
[29] P. Isola, J.-Y. Zhu, T. Zhou, A. A. Efros, and B. A. Research, “Image-to-Image Translation with Conditional Adversarial Networks.” [Online]. Available: https://github.com/phillipi/pix2pix.
[30] U. Demir and G. Unal, “Patch-Based Image Inpainting with Generative Adversarial Networks,” 2018, [Online]. Available: http://arxiv.org/abs/1803.07422.
[31] J. T. Zhou et al., “Sc2Net: Sparse LSTMs for sparse coding ∗,” 32nd AAAI Conf. Artif. Intell. AAAI 2018, pp. 4588–4595, 2018.
[32] T. A. Koleck, C. Dreisbach, P. E. Bourne, and S. Bakken, “Natural language processing of symptoms documented in free-text narratives of electronic health records: A systematic review,” J. Am. Med. Informatics Assoc., vol. 26, no. 4, pp. 364–379, 2019, doi: 10.1093/jamia/ocy173.
[33] Q.-Y. Zhong et al., “Use of natural language processing in electronic medical records to identify pregnant women with suicidal behavior: towards a solution to the complex classification problem,” Eur. J. Epidemiol., vol. 34, no. 2, pp. 153–162, Feb. 2019, doi: 10.1007/s10654-018-0470-0.
[34] C. Cortes and V. Vapnik, “Support-Vector Networks,” Mach. Learn., vol. 20, no. 3, pp. 273–297, Sep. 1995, doi: 10.1023/A:1022627411411.
[35] G. Varoquaux, L. Buitinck, G. Louppe, O. Grisel, F. Pedregosa, and A. Mueller, “Scikit-learn,” GetMobile Mob. Comput. Commun., vol. 19, no. 1, pp. 29–33, 2015, doi: 10.1145/2786984.2786995.
[36] L. Breiman, “Random Forests,” Mach. Learn., vol. 45, no. 1, pp. 5–32, Oct. 2001, doi: 10.1023/A:1010933404324.
[37] A. M. Kibriya, E. Frank, B. Pfahringer, and G. Holmes, “Multinomial Naive Bayes for Text Categorization Revisited,” in Proceedings of the 17th Australian Joint Conference on Advances in Artificial Intelligence, 2004, pp. 488–499, doi: 10.1007/978-3-540-30549-1_43.
[38] T. Mikolov, S. Kombrink, L. Burget, J. Černocký, and S. Khudanpur, “Extensions of recurrent neural network language model,” in ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings, 2011, pp. 5528–5531, doi: 10.1109/ICASSP.2011.5947611.
[39] S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Comput., vol. 9, no. 8, pp. 1735–1780, 1997, doi: 10.1162/neco.1997.9.8.1735.
[40] A. Graves, A. Mohamed, and G. Hinton, “Speech recognition with deep recurrent neural networks,” in 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, 2013, pp. 6645–6649, doi: 10.1109/ICASSP.2013.6638947.
[41] M. Sundermeyer, R. Schlüter, and H. Ney, “LSTM neural networks for language modeling,” 13th Annu. Conf. Int. Speech Commun. Assoc. 2012, INTERSPEECH 2012, vol. 1, pp. 194–197, 2012.
[42] I. Sutskever, O. Vinyals, and Q. V. Le, “Sequence to sequence learning with neural networks,” Adv. Neural Inf. Process. Syst., vol. 4, no. January, pp. 3104–3112, 2014.
[43] X. Yuan, L. Li, and Y. Wang, “Nonlinear Dynamic Soft Sensor Modeling with Supervised Long Short-Term Memory Network,” IEEE Trans. Ind. Informatics, vol. 16, no. 5, pp. 3168–3176, 2020, doi: 10.1109/TII.2019.2902129.
[44] L. Van Der Maaten and G. Hinton, “Visualizing Data using t-SNE,” 2008.
[45] J. D. Hunter, “Matplotlib: A 2D graphics environment,” Comput. Sci. Eng., vol. 9, no. 3, pp. 99–104, May 2007, doi: 10.1109/MCSE.2007.55.
[46] M. Waskom et al., “mwaskom/seaborn: v0.8.1 (September 2017).” Zenodo, Sep. 2017, doi: 10.5281/zenodo.883859.
[47] T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016, pp. 785–794, doi: 10.1145/2939672.2939785.
[48] W. Mckinney, “Data Structures for Statistical Computing in Python,” 2010.
[49] D. E. Rumelhart, G. E. Hinton, and R. J. Williams, “Learning Internal Representations by Error Propagation,” in Parallel Distributed Processing: Explorations in the Microstructure of Cognition, Vol. 1: Foundations, Cambridge, MA, USA: MIT Press, 1986, pp. 318–362.
電子全文 電子全文(網際網路公開日期:20250812)
連結至畢業學校之論文網頁點我開啟連結
註: 此連結為研究生畢業學校所提供,不一定有電子全文可供下載,若連結有誤,請點選上方之〝勘誤回報〞功能,我們會盡快修正,謝謝!
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
無相關期刊