Premature ventricular contraction beat detection with deep neural networks

Size: px
Start display at page:

Download "Premature ventricular contraction beat detection with deep neural networks"

Transcription

1 th IEEE International Conference on Machine Learning and Applications Premature ventricular contraction beat detection with deep neural networks Tae Joon Jun, Hyun Ji Park Nguyen Hoang Minh, Daeyoung Kim School of Computing KAIST Daejeon, Korea {taejoon89, hyunjip, minhhoang, Young-Hak Kim Department of Medicine University of Ulsan College of Medicine Ulsan, Korea Abstract A deep neural networks is proposed for the classification of premature ventricular contraction (PVC) beat, which is an irregular heartbeat initiated by Purkinje fibers rather than by sinoatrial node. Several machine learning approaches were proposed for the detection of PVC beats although they resulted in either achieving low accuracy of classification or using limited portion of data from existing electrocardiography (ECG) databases. In this paper, we propose an optimized deep neural networks for PVC beat classification. Our method is evaluated on TensorFlow, which is an open source machine learning platform initially developed by Google. Our method achieved overall 99.41% accuracy and a sensitivity of 96.08% with total 80,836 ECG beats including normal and PVC from the MIT-BIH Arrhythmia Database. Keywords Arrhythmia; Premature ventricular contraction; Deep neural network; Machine learning; TensorFlow I. INTRODUCTION Cardiovasular diseases (CVDs) are the number one cause of death in the world especially the main reason of death in developed countries. According to the World Health Organization (WHO), over 17.5 million people died from CVDs in 2012, which is more than 30% of all deaths in the world. However, it is known that people with CVDs or at high risk of CVDs can be managed by early detection of the disease and appropriate treatments. Therefore, the detection of CVDs for prevention and treatment is a significant task in medical health domain. There are several types of CVDs including heart valve problems, arrhythmia, and coronary artery disease. Detection of premature ventricular contractions (PVCs) is one of particular interest in case of arrhythmia. The PVC, also known as ventricular premature beat (VPB) is an irregular heartbeat started from ventricles while normal beat is initiated by sinoatrial node. A single PVC beat is commonly detected from healthy people and usually skipped without any symptoms. However, continuous PVC beats occasionally become a ventricular tachycardia (VT) that is a potentially fatal arrhythmia. Even worse, PVC beats become a ventricular fibrillation (VF), which can immediately lead people to death. Therefore detection of PVC beats are important for the patient with CVDs. Various techniques have been proposed for the PVC beat detection over the past several years. Machine learning algorithms such as Hidden Markov Chain [1], [2], principal component analysis [3], support vector machine [4], [5], [6], K-nearest neighbor classification [7], [8], Bayesian filters [9], and neural networks [10], [11] had been applied for the classification of arrhythmia beats including PVCs. Although these approaches have reached auspicious results, they have several limitations. First, the vast majority of these methods only applied in small subset of ECG database. In case of methods using MIT-BIH Arrhythmia database, although there are more than 80,000 ECG beats including normal and PVC beats, most of the previous researches retrieved small subset of data. Understandably, these methods reached extremely high accuracy such as over 99%. We do understand these methods are appropriate for the patient-to-patient PVC detection such as when using Holter recording. However, we do not believe that these approaches are suitable for today s general machine learning tendency especially related with Big data analysis. Second, there are methods that used entire MIT-BIH database for the classification, although they resulted in relatively poor accuracy and sensitivity. In this paper, we propose a highly accurate strategy for classifying PVC beats using deep neural networks that contains multiple hidden layers with 6 different input features extracted from ECG signals. The proposed neural networks is carefully optimized in selection of pre-trained dataset, input feature combinations, k value in k-fold cross validation, hidden layer architecture, regularization methods, the number of training steps, learning rates, activation functions, and optimizer functions. The proposed strategy achieved overall 99.41% accuracy and a sensitivity of 96.08% with total 80,835 ECG beats including normal and PVC from the MIT-BIH Arrhythmia database. As far as we know, our PVC detection method is the most accurate strategy in case of using total patient dataset. The rest of this paper is organized as follows. Section 2 presents the detailed explanation of the proposed classification strategy. Experimental evaluation and results are shown in Section 3 and 4. The conclusion of this paper is presented in Section /16 $ IEEE DOI /ICMLA

2 II. METHODOLOGY Our PVC detection strategy is composed of two main phases, feature extraction and beat classification. Fig. 1 presents the flow diagram of proposed PVC classification. Purpose of feature extraction phase is to detect a single heart beat pattern from sequential ECG signals. In general, combination of three different characteristic points called QRS complex is detected. This QRS complex contains important features that can be used to separate different heart beat types. In our detection method, we adopted well known Pan- Tompkins algorithm [12] to detect 4 different feature points; R-peak amplitude (R-peak), RR interval time (RR), QRSduration time (QRS), and ventricular activation time (VAT). In addition, we extracted two more significant features called Q-peak and S-peak amplitude. Fig. 2 describes 6 features extracted from the ECG signal. 6 features extracted from the first phase are used as inputs of beat classification phase. We built a deep neural network with 6 nodes of input layer, multiple hidden layers, and 2 nodes of output layer. A. Feature Extraction Before the features are extracted from sequential ECG signals, Pan-Tompkins algorithm performs 4 major QRS complex detection steps: band-pass filtering, five-point derivation, squaring, and moving window integration. Additionally, we added 5th step called peak-point detection after the moving window integration step to detect the Q-peak and S-peak. 1) Band-pass filter: Since the sequential ECG signal includes several noises such as muscle noise, power line interference, and baseline wandering, band-pass filtering eliminates these noises with combination of low-pass filter and highpass filter. The transfer function used for the low-pass filter is presented in following: H(z) = (1 z 6 ) 2 (1 z 1 ) 2 (1) When T is sampling rate of ECG signals, amplitude response function for the low-pass filter is shown in following: Fig. 2. Extracted features from the ECG signal H(wT) = sin2 (3ωT) sin 2 (2) (ωt/2) Difference equation for eliminating frequency higher than 11Hz is given in following: y(nt )=2y(nT T ) y(nt 2T )+x(nt ) (3) 2x(nT 6T )+x(nt 12T ) After the ECG signal pass through the low-pass filter, it enters the high-pass filter with transfer function presented in following: H(z) = ( 1+32z 16 + z 32 ) 1+z 1 (4) Similar to equation (2), amplitude response function for the high-pass filter is shown in following: [ sin 2 (16ωT) ] 1/2 H(wT) = (5) cos(wt/2) Difference equation for eliminating frequency lower than 5Hz is given in following: y(nt )=32x(nT 16T ) [y(nt T ) (6) +x(nt ) x(nt 32T )] 2) Five-point derivation: Subsequently, the five-point derivation is carried out to calculate gradients of the filtered ECG signals. Transfer function for the five-point derivation is presented in following: H(z) = 1 8T ( z 2 2z 1 +2z 1 + z 2 ) (7) Amplitude response function and difference equation for the derivation are shown in following: H(wT) = 1 [sin(2ωt)+2sin(ωt)] (8) 4T Fig. 1. PVC detection flow diagram y(nt )= 1 [ x(nt 2T ) 2x(nT T ) 8T +2x(nT + T )+x(nt +2T )] (9) 860

3 3) Squaring: After the five-point derivation, QRS waves of the ECG signal are roughly detected. Through out the squaring phase, ECG signal is converted into positive value and the QRS wave signal is accentuated where the frequencies are relatively higher than others. Difference equation for the squaring is given in following: y(nt )=[x(nt )] 2 (10) 4) Moving window integration: Moving window integration is the last phase of the Pan-Tompkins algorithm. Through out the integration, we can detect starting point of Q wave, peak point of R wave, and end point of S wave. When N is the number of samples in the moving window, equation for the integration is presented in following: y(nt )= 1 [x(nt (N 1)T )+x(nt N (N 2)T ) x(nt )] (11) 5) Peak-point detection: After the moving window integration phase, the three points of the QRS complex are detected; Q wave starting time, R wave peak time, and S wave end time. Detection of the Q-peak and S-peak uses these points as a time stamp. When the window approaches Q wave starting time at the moving window integration step, we set the Q-flag value to true. From the Q wave starting point to the R wave peak point, minimum amplitude and time from the ECG signal are stored. Note that these amplitude information is obtained from the filtered ECG signal not from the raw signal. When the window arrives at R wave peak time, we retrieve the stored value that is Q-peak amplitude and time. After the R wave peak time, we set the Q-flag value to false and refresh the stored minimum value and time. Repeating the similar method, we obtain S- peak amplitude and time between the R wave peak point and the S wave end point, B. Beat Classification Deep learning and deep neural networks are considered as the main machine learning techniques for the proposed PVC beat classification. The main drawback of the traditional neural networks was that when the number of the hidden layers and nodes increased, the model itself easily fell into local minimum point. In 2006, Hinton et al. dramatically solved this problem with pre-training each layer with unsupervised learning [13]. After Hinton s approach, deep neural networks emerged on the horizon with Big data era and General Purpose Graphic Processing Unit technique (GPGPU). Optimization of the proposed deep neural networks considered several parameters; K value in K-fold cross validation, combination of 6 features, hidden layer architecture, Xavier weight initialization method, regularization methods, the number of training steps, learning rates, activation function types, and optimizer function types. 1) Data normalization: Before starting the optimization in earnest, we normalized extracted feature data with Min-Max normalization as data pre-processing step. x = x min (12) Max min 2) Feature combination: The 6 extracted features represent the key characteristics of ECG beat type. To figure out the best feature set, we combined four to six different number of set with every 6 feature. This is the the most important step in the optimization since the domain experts classifies ECG beat types by checking distinguishing points, which are features, from the signal. As we later seen from the evaluation results, decision of which features as input set conclusively affects the overall PVC detection accuracy. 3) K-fold cross validation: K-fold cross validation is a model validation technique for enhancing the accuracy of the statistical analysis when the data set is relatively small. When there is N number of data set, it is divided into K different subsets. We leave the first subset as test set and train the model with remaining K - 1 subsets. After repeatedly performing this step with K different subsets as test set, the overall accuracy of the model is calculated with average accuracy of K different test sets. Fig. 3 describes the overall procedure of K-fold cross validation. Optimization required in this step is to find the most appropriate K value that minimizes over-fitting problem while presents reasonable average accuracy. 4) Hidden layer architecture: After deciding input features and K value, constructing hidden layer architecture is required. In this step, we consider the number of hidden layers and the number of nodes in each layer. Starting from single hidden layer, we continuously increased and modified the number of hidden layers and nodes until the overall accuracy saturates. As result, we constructed 6 hidden layers with L2 and dropout regularization with dropout rate 0.9 [14]. Fig. 4 briefly shows our 6 hidden layer architecture. 5) Xavier initialization: Xavier et al. introduced empirical initialization method for the weights of a matrix [15]. This initialization technique suggests that for the sigmoid and tanh activation functions, model can achieve better accuracy and faster convergence if the weights are initialized randomly with the following range: Fig. 3. K-fold cross validation 861

4 Fig Hidden Layer DNN Architecture x = 6 ;[ x, x] (13) In + Out Where In is the number of input units to the weights and Out is the number of output units. 6) Regularization: The purpose of the regularization in neural networks is to prevent over-fitting of the model by regulating capacity of the networks. We applied L2 regularization to weight vectors not to bias ones. When the number of the hidden layer reaches 5, our model seriously fell into local minimum problem. To prevent both over-fitting and local minimum problems, we applied dropout technique. Selection of dropout rate and layers to apply dropout was important in this phase. Especially, dropout shouldn t be applied to the first hidden layer in the neural networks since our input set only consists maximum to 6 features. when there was a loss from the first layer, overall accuracy of PVC detection dramatically decreased. 7) Training steps: In general, the number of training steps affects the neural networks with over-fitting problem. If the number of training steps is too small, procedure would be terminated without sufficient training, which results poor overall accuracy. In contrast, if the number of training steps is exceedingly large, the model will fell into over-fitting problem which also results in relatively low accuracy. Popular way to find the best training steps is using validation set. When the accuracy of validation set starts to decrease, we stop our model from training and applied it to test set. However, our empirical results supports that well optimized neural networks does not easily fell into over-fitting problem regardless of the number of training steps increases. Therefore, we pointed out some typical number of training steps in evaluation step between from 10 6 to check the relationship between the classification efficiency and the number of training steps. 8) Learning rate: We initialized learning rate with 0.01, since empirically we noticed that this rate works relatively fine. However, we also concluded that initial learning rate itself does not affects the performance since even large initial learning rate, we can manage the cost minimization by applying an exponential decay function to decrease the learning rate as the training continuous. As result, we set initial learning rate to 0.01 and exponential decay rate 0.95 with every 10,000 training steps. 9) Activation function: Activation function defines the output of nodes from given inputs in neural networks. Unlike traditional neural networks, deep learning technique normally handles normalized non-linear activation functions to prohibit inputs of node increasing with out bound. We compared several non-linear activation functions in this work. a) Sigmoid: Sigmoid function is a typical non-linear activation function in neural networks and it is a special case of the logistic function. 1 sigmoid(t) = 1+e t (14) b) Hyperbolic tangent: The hyperbolic tangent (tanh) function is frequently used when negative output value of the activation function is required. Different from the sigmoid function, hyperbolic tangent function maps the inputs between -1 and 1. tanh(x) = sinh(x) consh(x) = 1 e 2x 1+e 2x (15) c) Rectified linear unit: The rectified linear unit (ReLU) function is the most popular activation function in deep neural networks especially when using convolutional neural networks (CNN). relu(x) =max(0,x) (16) d) Exponential linear unit: Clevert et al. introduced exponential linear unit (ELU) in [16]. ELU reduces bias shift effect by allowing negative values while presents relatively well performance compared to ReLU. { x if x < 0 elu(x) = (17) α(exp(x) 1) if x 0 e) Softplus: The softplus function is a smooth approximation to the ReLU. Since the ReLU is not differentiable function, softplus tried to follow the characteristics of ReLU while presenting derivatives at every point. softplus(x) =ln(1 + e x ) (18) f) Softsign: The softsign function can be classified similar to sigmoid and hyperbolic tangent functions. This function provides near zero unit average while prevents function from saturating easily. softsign(x) = z (19) 1+ z 10) Optimizer function: Optimizer function is required to minimize the cost which means difference between actual label values and estimated values. After we calculated softmax cross entropy between estimated value and result value, we compared accuracy over different optimizer functions such as gradient descent, Adam[17], Adagrad[18], and RMSProp[19]. 862

5 III. EVALUATION The proposed PVC classification has been evaluated to validate its overall accuracy and sensitivity with various parameters we introduced in section 3. The measures to describe the accuracy and sensitivity were: a) True Positive (TP): Correctly detected as PVC b) True Negative (TN): Correctly detected as normal c) False Positive (FP): Incorrectly detected as PVC d) False Negative (FN): Incorrectly detected as normal Using these measurements, overall accuracy and sensitivity can be calculated by following equations: Accuracy(%) = TP + TN 100 (20) TP + TN + FP + FN Sensitivity(%) = TP TP + FP 100 (21) The raw ECG signal was obtained from the MIT-BIH Arrhythmia database. In this database, the total number of ECG beats are 109,494. However, from the feature extraction phase, several beats are lost or unclassified as ECG beats. Therefore, we detected 108,022 ECG beats from the database using Pan-Tompkins algorithms which is 98.66% detection accuracy for QRS complex. After the 6 features are extracted, we excluded ECG beats which are neither normal nor PVC. As result, we retrieved 80,836 ECG beats including 74,157 normal beats and 6,678 PVC beats. Next we separated entire data set into training set and test set. Distribution of data was uniformly processed by using divisor and remainder. The implementation of deep neural networks was performed in TensorFlow[20] with GPGPU support. Since we applied K-fold cross validation, total execution time for the training requires K times more than the naive one. Therefore, we used 4 NVIDIA K20m GPUs to accelerate the training execution time. IV. RESULT This section provides the results of the accuracy and sensitivity with optimization steps we took in section 3. Experiments processed in three steps: A. Experiment 1 In this experiment, we optimized simple deep neural networks which has 3 hidden layers with different K values from 3 to 20, and with different feature combinations to find best feature set and K-value. Using entire features is named FC1, and accounting 5 features are named FC2 to FC7. Since the overall accuracy results were around 99% in every experiments, we only presents the sensitivity results for the Experiment 1. Name FC1 FC2 FC3 FC4 FC5 FC6 FC7 TABLE I FEATURE COMBINATIONS Features Rpeak RR QRS VAT Qpeak Speak Rpeak RR QRS VAT Qpeak Rpeak QRS VAT Qpeak Speak Rpeak RR VAT Qpeak Speak Rpeak RR QRS Qpeak Speak Rpeak RR QRS VAT Speak RR QRS VAT Qpeak Speak TABLE II EXPERIMENT 1RESULT -SENSITIVITY(%) K FC1 FC2 FC3 FC4 FC5 FC6 FC B. Experiment 2 From the previous experiment, K = 8with FC1 presented best result. In this experiment, we optimized deep neural networks with different number of hidden layers and number of nodes in each layer. Starting from single hidden layer, we increased the number of layers until it reached 4. Although the accuracy and sensitivity keep increased, training often fell into local minimum point. After 5 hidden layers, most of the training tumbled to local minimum which results sensitivity near zero. Therefore, starting from 5 hidden layers, we applied L2 regularization and dropout with rate 0.9 to the weight matrix and nodes. Especially in 3 hidden layers experiment, we evaluated three different types of hidden layer architecture with the same number of nodes: diamond shape, increasing triangle shape, decreasing triangle shape. Fig. 5 demonstrated these architectural shapes. Between these three shapes, we concluded that diamond shape works better than other two based on experimental results. TABLE III EXPERIMENT 2RESULT L1 L2 L3 L4 L5 L6 L7 Acc(%) Se(%) Fig. 5. Hidden Layer Architectural Shapes 863

6 C. Experiment 3 In the Experiment 2, the best result was experiment with 6 hidden layers using L2 regularization and dropout rate of 0.9 except the first hidden layer. The total number of nodes we used in 6 hidden layers are 10 x 30 x 50 x 100 x 40 x 20. In third experiment, we compared the result of different training steps to figure out minimum training steps required to obtain reasonable boundary of results. Fig. 6 displays the accuracy and sensitivity with different number of training steps. D. Experiment 4 In this experiment, we compared the result of different activation and optimizer functions to tune the neural networks obtained from previous experiments. V. DISCUSSION From the experiments, we achieved the overall accuracy of 99.41% and sensitivity of 96.08%. To achieve this accuracy, we used Min-Max normalization, L2 regularization, dropout with rate of 0.9, 6 hidden layer DNN with 100,000 training steps, 0.01 initial learning rate with 0.95 exponent dacay funtion every 10,000 training steps, Xavier weight initialization, Softsign activation function, and Adam optimizer function. As far as we know, our proposed classification method provides the highest accuracy and sensitivity with MIT-BIH Arrhythmia database. Although, some of the related researches achieved higher results with partial set of database or patient-to-patient approach, we believe that using entire data set as training and test set is more suitable for recent machine learning applications, since they require enormous size of data for training. Fig. 6. Experiment 3 Result TABLE IV EXPERIMENT 4RESULT -ACTIVATION Softplus Relu Elu Tanh Softsign Sigmoid Acc(%) Se(%) TABLE V EXPERIMENT 4RESULT -OPTIMIZER Adam Adagrad RMSProp GradientDescent Acc(%) Se(%) VI. CONCLUSION In this paper, we propose an optimized deep neural networks for PVC beat classification. Our method achieved overall 99.41% accuracy and a sensitivity of 96.08%. We believe our deep neural networks model can be applied to other medical health databases and looking forward to evaluated them as future works. REFERENCES [1] W. T. Cheng and K. L. Chan, Classification of electrocardiogram using hiddenmarkovmodels in Proc. 20th Annu. Int. Conf. IEEE EMBS, 1998, vol. 20, pp [2] D. A. Coast, R. M. Stern, G. G. Cano, and S. A. Briller, An approach to Cardiac arrhythmia analysis using hidden Markov models IEEE Trans. Biomed. Eng., vol. 37, no. 9, pp , Sep [3] G. B. Moody and R. G. Mark, QRS morphology representation and noise estimation using the KarhunenLo eve transform in Proc. Comput. Cardiol., 1989, pp [4] Melgani, Farid, and Yakoub Bazi. Classification of electrocardiogram signals with support vector machines and particle swarm optimization in IEEE Transactions on Information Technology in Biomedicine 12.5 (2008): [5] Faziludeen, Shameer, and P. V. Sabiq. ECG beat classification using wavelets and SVM in Information & Communication Technologies (ICT), 2013 IEEE Conference on. IEEE, [6] Nuryani, Nuryani, Iwan Yahya, and Anik Lestari. Premature ventricular contraction detection using swarm-based support vector machine and QRS wave features in International Journal of Biomedical Engineering and Technology 16.4 (2014): [7] I. Christov, I. Jekova, and G. Bortolan, Premature ventricular contraction classification by the Kth nearest-neighbours rule, in Physiol. Meas., vol. 26, pp , [8] Park, Juyoung, Kuyeon Lee, and Kyungtae Kang. Arrhythmia detection from heartbeat using k-nearest neighbor classifier, Bioinformatics and Biomedicine (BIBM), 2013 IEEE International Conference on. IEEE, [9] Sayadi, Omid, Mohammad B. Shamsollahi, and Gari D. Clifford. Robust detection of premature ventricular contractions using a wave-based Bayesian framework in IEEE Transactions on Biomedical Engineering 57.2 (2010): [10] Wang, Jeen-Shing, et al. ECG arrhythmia classification using a probabilistic neural network with a feature reduction method in Neurocomputing 116 (2013): [11] Javadi, Mehrdad, et al. Classification of ECG arrhythmia by a modular neural network based on mixture of experts and negatively correlated learning in Biomedical Signal Processing and Control 8.3 (2013): [12] Pan, Jiapu, and Willis J. Tompkins. A real-time QRS detection algorithm in IEEE transactions on biomedical engineering 3 (1985): [13] Hinton, Geoffrey E., Simon Osindero, and Yee-Whye Teh. A fast learning algorithm for deep belief nets in Neural computation 18.7 (2006): [14] Srivastava, Nitish, et al. Dropout: a simple way to prevent neural networks from overfitting in Journal of Machine Learning Research 15.1 (2014): [15] Glorot, Xavier, and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks in Aistats. Vol [16] Clevert, Djork-Arn, Thomas Unterthiner, and Sepp Hochreiter. Fast and accurate deep network learning by exponential linear units (elus) in arxiv preprint arxiv: (2015). [17] Kingma, Diederik, and Jimmy Ba. Adam: A method for stochastic optimization in arxiv preprint arxiv: (2014). [18] Duchi, John, Elad Hazan, and Yoram Singer. Adaptive subgradient methods for online learning and stochastic optimization in Journal of Machine Learning Research 12.Jul (2011): [19] Tieleman, Tijmen, and Geoffrey Hinton. Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude in COURSERA: Neural Networks for Machine Learning 4.2 (2012). [20] Abadi, Martn, et al. TensorFlow: Large-scale machine learning on heterogeneous systems, 2015 Software available from tensorflow. org 1 (2015). 864

Deep Learning with Tensorflow AlexNet

Deep Learning with Tensorflow   AlexNet Machine Learning and Computer Vision Group Deep Learning with Tensorflow http://cvml.ist.ac.at/courses/dlwt_w17/ AlexNet Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton, "Imagenet classification

More information

Arrhythmia Classification via k-means based Polyhedral Conic Functions Algorithm

Arrhythmia Classification via k-means based Polyhedral Conic Functions Algorithm Arrhythmia Classification via k-means based Polyhedral Conic Functions Algorithm Full Research Paper / CSCI-ISPC Emre Cimen Anadolu University Industrial Engineering Eskisehir, Turkey ecimen@anadolu.edu.tr

More information

Natural Language Processing CS 6320 Lecture 6 Neural Language Models. Instructor: Sanda Harabagiu

Natural Language Processing CS 6320 Lecture 6 Neural Language Models. Instructor: Sanda Harabagiu Natural Language Processing CS 6320 Lecture 6 Neural Language Models Instructor: Sanda Harabagiu In this lecture We shall cover: Deep Neural Models for Natural Language Processing Introduce Feed Forward

More information

ECG Arrhythmia Classification using Least Squares Twin Support Vector Machines

ECG Arrhythmia Classification using Least Squares Twin Support Vector Machines 26th Iranian Conference on Electrical Engineering (ICEE2018) ECG Arrhythmia Classification using Least Squares Twin Support Vector Machines Mohammad S. Refahi Department of Electrical Engineering Amirkabir

More information

Neural Network Optimization and Tuning / Spring 2018 / Recitation 3

Neural Network Optimization and Tuning / Spring 2018 / Recitation 3 Neural Network Optimization and Tuning 11-785 / Spring 2018 / Recitation 3 1 Logistics You will work through a Jupyter notebook that contains sample and starter code with explanations and comments throughout.

More information

Study of Residual Networks for Image Recognition

Study of Residual Networks for Image Recognition Study of Residual Networks for Image Recognition Mohammad Sadegh Ebrahimi Stanford University sadegh@stanford.edu Hossein Karkeh Abadi Stanford University hosseink@stanford.edu Abstract Deep neural networks

More information

Clustering Of Ecg Using D-Stream Algorithm

Clustering Of Ecg Using D-Stream Algorithm Clustering Of Ecg Using D-Stream Algorithm Vaishali Yeole Jyoti Kadam Department of computer Engg. Department of computer Engg. K.C college of Engg, K.C college of Engg Thane (E). Thane (E). Abstract The

More information

Deep Learning Applications

Deep Learning Applications October 20, 2017 Overview Supervised Learning Feedforward neural network Convolution neural network Recurrent neural network Recursive neural network (Recursive neural tensor network) Unsupervised Learning

More information

Stacked Denoising Autoencoders for Face Pose Normalization

Stacked Denoising Autoencoders for Face Pose Normalization Stacked Denoising Autoencoders for Face Pose Normalization Yoonseop Kang 1, Kang-Tae Lee 2,JihyunEun 2, Sung Eun Park 2 and Seungjin Choi 1 1 Department of Computer Science and Engineering Pohang University

More information

Index. Umberto Michelucci 2018 U. Michelucci, Applied Deep Learning,

Index. Umberto Michelucci 2018 U. Michelucci, Applied Deep Learning, A Acquisition function, 298, 301 Adam optimizer, 175 178 Anaconda navigator conda command, 3 Create button, 5 download and install, 1 installing packages, 8 Jupyter Notebook, 11 13 left navigation pane,

More information

Convolutional Neural Networks

Convolutional Neural Networks Lecturer: Barnabas Poczos Introduction to Machine Learning (Lecture Notes) Convolutional Neural Networks Disclaimer: These notes have not been subjected to the usual scrutiny reserved for formal publications.

More information

Real-time convolutional networks for sonar image classification in low-power embedded systems

Real-time convolutional networks for sonar image classification in low-power embedded systems Real-time convolutional networks for sonar image classification in low-power embedded systems Matias Valdenegro-Toro Ocean Systems Laboratory - School of Engineering & Physical Sciences Heriot-Watt University,

More information

Lecture 20: Neural Networks for NLP. Zubin Pahuja

Lecture 20: Neural Networks for NLP. Zubin Pahuja Lecture 20: Neural Networks for NLP Zubin Pahuja zpahuja2@illinois.edu courses.engr.illinois.edu/cs447 CS447: Natural Language Processing 1 Today s Lecture Feed-forward neural networks as classifiers simple

More information

Deep Learning. Practical introduction with Keras JORDI TORRES 27/05/2018. Chapter 3 JORDI TORRES

Deep Learning. Practical introduction with Keras JORDI TORRES 27/05/2018. Chapter 3 JORDI TORRES Deep Learning Practical introduction with Keras Chapter 3 27/05/2018 Neuron A neural network is formed by neurons connected to each other; in turn, each connection of one neural network is associated

More information

Lecture : Neural net: initialization, activations, normalizations and other practical details Anne Solberg March 10, 2017

Lecture : Neural net: initialization, activations, normalizations and other practical details Anne Solberg March 10, 2017 INF 5860 Machine learning for image classification Lecture : Neural net: initialization, activations, normalizations and other practical details Anne Solberg March 0, 207 Mandatory exercise Available tonight,

More information

Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah

Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah Improving the way neural networks learn Srikumar Ramalingam School of Computing University of Utah Reference Most of the slides are taken from the third chapter of the online book by Michael Nielson: neuralnetworksanddeeplearning.com

More information

CPSC 340: Machine Learning and Data Mining. Deep Learning Fall 2016

CPSC 340: Machine Learning and Data Mining. Deep Learning Fall 2016 CPSC 340: Machine Learning and Data Mining Deep Learning Fall 2016 Assignment 5: Due Friday. Assignment 6: Due next Friday. Final: Admin December 12 (8:30am HEBB 100) Covers Assignments 1-6. Final from

More information

Advanced Introduction to Machine Learning, CMU-10715

Advanced Introduction to Machine Learning, CMU-10715 Advanced Introduction to Machine Learning, CMU-10715 Deep Learning Barnabás Póczos, Sept 17 Credits Many of the pictures, results, and other materials are taken from: Ruslan Salakhutdinov Joshua Bengio

More information

LOW POWER FPGA IMPLEMENTATION OF REAL-TIME QRS DETECTION ALGORITHM

LOW POWER FPGA IMPLEMENTATION OF REAL-TIME QRS DETECTION ALGORITHM LOW POWER FPGA IMPLEMENTATION OF REAL-TIME QRS DETECTION ALGORITHM VIJAYA.V, VAISHALI BARADWAJ, JYOTHIRANI GUGGILLA Electronics and Communications Engineering Department, Vaagdevi Engineering College,

More information

Computer Vision: Homework 5 Optical Character Recognition using Neural Networks

Computer Vision: Homework 5 Optical Character Recognition using Neural Networks 16-720 Computer Vision: Homework 5 Optical Character Recognition using Neural Networks Instructors: Deva Ramanan TAs: Achal Dave*, Sashank Jujjavarapu, Siddarth Malreddy, Brian Pugh Originally developed

More information

An Introduction to NNs using Keras

An Introduction to NNs using Keras An Introduction to NNs using Keras Michela Paganini michela.paganini@cern.ch Yale University 1 Keras Modular, powerful and intuitive Deep Learning python library built on Theano and TensorFlow Minimalist,

More information

Machine Learning. Deep Learning. Eric Xing (and Pengtao Xie) , Fall Lecture 8, October 6, Eric CMU,

Machine Learning. Deep Learning. Eric Xing (and Pengtao Xie) , Fall Lecture 8, October 6, Eric CMU, Machine Learning 10-701, Fall 2015 Deep Learning Eric Xing (and Pengtao Xie) Lecture 8, October 6, 2015 Eric Xing @ CMU, 2015 1 A perennial challenge in computer vision: feature engineering SIFT Spin image

More information

Neural Networks for Machine Learning. Lecture 15a From Principal Components Analysis to Autoencoders

Neural Networks for Machine Learning. Lecture 15a From Principal Components Analysis to Autoencoders Neural Networks for Machine Learning Lecture 15a From Principal Components Analysis to Autoencoders Geoffrey Hinton Nitish Srivastava, Kevin Swersky Tijmen Tieleman Abdel-rahman Mohamed Principal Components

More information

Research on Pruning Convolutional Neural Network, Autoencoder and Capsule Network

Research on Pruning Convolutional Neural Network, Autoencoder and Capsule Network Research on Pruning Convolutional Neural Network, Autoencoder and Capsule Network Tianyu Wang Australia National University, Colledge of Engineering and Computer Science u@anu.edu.au Abstract. Some tasks,

More information

Deep Learning. Volker Tresp Summer 2014

Deep Learning. Volker Tresp Summer 2014 Deep Learning Volker Tresp Summer 2014 1 Neural Network Winter and Revival While Machine Learning was flourishing, there was a Neural Network winter (late 1990 s until late 2000 s) Around 2010 there

More information

Object Detection Lecture Introduction to deep learning (CNN) Idar Dyrdal

Object Detection Lecture Introduction to deep learning (CNN) Idar Dyrdal Object Detection Lecture 10.3 - Introduction to deep learning (CNN) Idar Dyrdal Deep Learning Labels Computational models composed of multiple processing layers (non-linear transformations) Used to learn

More information

CS489/698: Intro to ML

CS489/698: Intro to ML CS489/698: Intro to ML Lecture 14: Training of Deep NNs Instructor: Sun Sun 1 Outline Activation functions Regularization Gradient-based optimization 2 Examples of activation functions 3 5/28/18 Sun Sun

More information

Cache-efficient Gradient Descent Algorithm

Cache-efficient Gradient Descent Algorithm Cache-efficient Gradient Descent Algorithm Imen Chakroun, Tom Vander Aa and Thomas J. Ashby Exascience Life Lab, IMEC, Leuven, Belgium Abstract. Best practice when using Stochastic Gradient Descent (SGD)

More information

Neural Networks. Single-layer neural network. CSE 446: Machine Learning Emily Fox University of Washington March 10, /10/2017

Neural Networks. Single-layer neural network. CSE 446: Machine Learning Emily Fox University of Washington March 10, /10/2017 3/0/207 Neural Networks Emily Fox University of Washington March 0, 207 Slides adapted from Ali Farhadi (via Carlos Guestrin and Luke Zettlemoyer) Single-layer neural network 3/0/207 Perceptron as a neural

More information

Deep Learning With Noise

Deep Learning With Noise Deep Learning With Noise Yixin Luo Computer Science Department Carnegie Mellon University yixinluo@cs.cmu.edu Fan Yang Department of Mathematical Sciences Carnegie Mellon University fanyang1@andrew.cmu.edu

More information

Lecture : Training a neural net part I Initialization, activations, normalizations and other practical details Anne Solberg February 28, 2018

Lecture : Training a neural net part I Initialization, activations, normalizations and other practical details Anne Solberg February 28, 2018 INF 5860 Machine learning for image classification Lecture : Training a neural net part I Initialization, activations, normalizations and other practical details Anne Solberg February 28, 2018 Reading

More information

URL: < z>

URL:   < z> Citation: Qin, Qin, Li, Jianqing, Zhang, Li, Yue, Yinggao and Liu, Chengyu (2017) Combining Low-dimensional Wavelet Features and Support Vector Machine for Arrhythmia Beat Classification. Scientific Reports,

More information

Design of a hybrid model for cardiac arrhythmia classification based on Daubechies wavelet transform

Design of a hybrid model for cardiac arrhythmia classification based on Daubechies wavelet transform Original papers Design of a hybrid model for cardiac arrhythmia classification based on Daubechies wavelet transform Rekha Rajagopal B F, Vidhyapriya Ranganathan A Department of Information Technology,

More information

COMP 551 Applied Machine Learning Lecture 16: Deep Learning

COMP 551 Applied Machine Learning Lecture 16: Deep Learning COMP 551 Applied Machine Learning Lecture 16: Deep Learning Instructor: Ryan Lowe (ryan.lowe@cs.mcgill.ca) Slides mostly by: Class web page: www.cs.mcgill.ca/~hvanho2/comp551 Unless otherwise noted, all

More information

JOINT INTENT DETECTION AND SLOT FILLING USING CONVOLUTIONAL NEURAL NETWORKS. Puyang Xu, Ruhi Sarikaya. Microsoft Corporation

JOINT INTENT DETECTION AND SLOT FILLING USING CONVOLUTIONAL NEURAL NETWORKS. Puyang Xu, Ruhi Sarikaya. Microsoft Corporation JOINT INTENT DETECTION AND SLOT FILLING USING CONVOLUTIONAL NEURAL NETWORKS Puyang Xu, Ruhi Sarikaya Microsoft Corporation ABSTRACT We describe a joint model for intent detection and slot filling based

More information

Optimizing the detection of characteristic waves in ECG based on processing methods combinations

Optimizing the detection of characteristic waves in ECG based on processing methods combinations Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000. Digital Object Identifier 10.1109/ACCESS.2017.Doi Number Optimizing the detection of characteristic waves in ECG based on processing

More information

Keywords: wearable system, flexible platform, complex bio-signal, wireless network

Keywords: wearable system, flexible platform, complex bio-signal, wireless network , pp.119-123 http://dx.doi.org/10.14257/astl.2014.51.28 Implementation of Fabric-Type Flexible Platform based Complex Bio-signal Monitoring System for Situational Awareness and Accident Prevention in Special

More information

Keras: Handwritten Digit Recognition using MNIST Dataset

Keras: Handwritten Digit Recognition using MNIST Dataset Keras: Handwritten Digit Recognition using MNIST Dataset IIT PATNA February 9, 2017 1 / 24 OUTLINE 1 Introduction Keras: Deep Learning library for Theano and TensorFlow 2 Installing Keras Installation

More information

Knowledge Discovery and Data Mining

Knowledge Discovery and Data Mining Knowledge Discovery and Data Mining Lecture 13 - Neural Nets Tom Kelsey School of Computer Science University of St Andrews http://tom.home.cs.st-andrews.ac.uk twk@st-andrews.ac.uk Tom Kelsey ID5059-13-NN

More information

2. Neural network basics

2. Neural network basics 2. Neural network basics Next commonalities among different neural networks are discussed in order to get started and show which structural parts or concepts appear in almost all networks. It is presented

More information

Introduction to Deep Learning

Introduction to Deep Learning ENEE698A : Machine Learning Seminar Introduction to Deep Learning Raviteja Vemulapalli Image credit: [LeCun 1998] Resources Unsupervised feature learning and deep learning (UFLDL) tutorial (http://ufldl.stanford.edu/wiki/index.php/ufldl_tutorial)

More information

DEEP LEARNING REVIEW. Yann LeCun, Yoshua Bengio & Geoffrey Hinton Nature Presented by Divya Chitimalla

DEEP LEARNING REVIEW. Yann LeCun, Yoshua Bengio & Geoffrey Hinton Nature Presented by Divya Chitimalla DEEP LEARNING REVIEW Yann LeCun, Yoshua Bengio & Geoffrey Hinton Nature 2015 -Presented by Divya Chitimalla What is deep learning Deep learning allows computational models that are composed of multiple

More information

Natural Language Processing with Deep Learning CS224N/Ling284. Christopher Manning Lecture 4: Backpropagation and computation graphs

Natural Language Processing with Deep Learning CS224N/Ling284. Christopher Manning Lecture 4: Backpropagation and computation graphs Natural Language Processing with Deep Learning CS4N/Ling84 Christopher Manning Lecture 4: Backpropagation and computation graphs Lecture Plan Lecture 4: Backpropagation and computation graphs 1. Matrix

More information

ELECTROCARDIOGRAM ABNORMALITY DETECTION BASED ON MULTIPLE MODELS. Gangyi Zhu, Congrong Guan, Ke Tan, Peidong Wang

ELECTROCARDIOGRAM ABNORMALITY DETECTION BASED ON MULTIPLE MODELS. Gangyi Zhu, Congrong Guan, Ke Tan, Peidong Wang ELECTROCARDIOGRAM ABNORMALITY DETECTION BASED ON MULTIPLE MODELS Gangyi Zhu, Congrong Guan, Ke Tan, Peidong Wang Department of Computer Science and Engineering The Ohio State University, Columbus, OH 4320.

More information

Deep Learning Based Real-time Object Recognition System with Image Web Crawler

Deep Learning Based Real-time Object Recognition System with Image Web Crawler , pp.103-110 http://dx.doi.org/10.14257/astl.2016.142.19 Deep Learning Based Real-time Object Recognition System with Image Web Crawler Myung-jae Lee 1, Hyeok-june Jeong 1, Young-guk Ha 2 1 Department

More information

An arrhythmia classification algorithm using a dedicated wavelet adapted to different subjects

An arrhythmia classification algorithm using a dedicated wavelet adapted to different subjects RESEARCH Open Access An arrhythmia classification algorithm using a dedicated wavelet adapted to different subjects Jinkwon Kim, Se Dong Min and Myoungho Lee * * Correspondence: ehealth@yonsei. ac.kr Department

More information

ImageNet Classification with Deep Convolutional Neural Networks

ImageNet Classification with Deep Convolutional Neural Networks ImageNet Classification with Deep Convolutional Neural Networks Alex Krizhevsky Ilya Sutskever Geoffrey Hinton University of Toronto Canada Paper with same name to appear in NIPS 2012 Main idea Architecture

More information

The exam is closed book, closed notes except your one-page cheat sheet.

The exam is closed book, closed notes except your one-page cheat sheet. CS 189 Fall 2015 Introduction to Machine Learning Final Please do not turn over the page before you are instructed to do so. You have 2 hours and 50 minutes. Please write your initials on the top-right

More information

Artificial Neural Networks (Feedforward Nets)

Artificial Neural Networks (Feedforward Nets) Artificial Neural Networks (Feedforward Nets) y w 03-1 w 13 y 1 w 23 y 2 w 01 w 21 w 22 w 02-1 w 11 w 12-1 x 1 x 2 6.034 - Spring 1 Single Perceptron Unit y w 0 w 1 w n w 2 w 3 x 0 =1 x 1 x 2 x 3... x

More information

Knowledge Discovery and Data Mining. Neural Nets. A simple NN as a Mathematical Formula. Notes. Lecture 13 - Neural Nets. Tom Kelsey.

Knowledge Discovery and Data Mining. Neural Nets. A simple NN as a Mathematical Formula. Notes. Lecture 13 - Neural Nets. Tom Kelsey. Knowledge Discovery and Data Mining Lecture 13 - Neural Nets Tom Kelsey School of Computer Science University of St Andrews http://tom.home.cs.st-andrews.ac.uk twk@st-andrews.ac.uk Tom Kelsey ID5059-13-NN

More information

A Fast Learning Algorithm for Deep Belief Nets

A Fast Learning Algorithm for Deep Belief Nets A Fast Learning Algorithm for Deep Belief Nets Geoffrey E. Hinton, Simon Osindero Department of Computer Science University of Toronto, Toronto, Canada Yee-Whye Teh Department of Computer Science National

More information

Deep Residual Architecture for Skin Lesion Segmentation

Deep Residual Architecture for Skin Lesion Segmentation Deep Residual Architecture for Skin Lesion Segmentation Venkatesh G M 1, Naresh Y G 1, Suzanne Little 2, and Noel O Connor 2 1 Insight Centre for Data Analystics-DCU, Dublin, Ireland 2 Dublin City University,

More information

Deep Neural Networks:

Deep Neural Networks: Deep Neural Networks: Part II Convolutional Neural Network (CNN) Yuan-Kai Wang, 2016 Web site of this course: http://pattern-recognition.weebly.com source: CNN for ImageClassification, by S. Lazebnik,

More information

QR CODE LOCALIZATION USING DEEP NEURAL NETWORKS

QR CODE LOCALIZATION USING DEEP NEURAL NETWORKS 2014 IEEE INTERNATIONAL WORKSHOP ON MACHINE LEARNING FOR SIGNAL PROCESSING, SEPT. 21 24, 2014, REIMS, FRANCE QR CODE LOCALIZATION USING DEEP NEURAL NETWORKS Tamás Grósz Péter Bodnár László Tóth László

More information

Deep Learning. Vladimir Golkov Technical University of Munich Computer Vision Group

Deep Learning. Vladimir Golkov Technical University of Munich Computer Vision Group Deep Learning Vladimir Golkov Technical University of Munich Computer Vision Group 1D Input, 1D Output target input 2 2D Input, 1D Output: Data Distribution Complexity Imagine many dimensions (data occupies

More information

Gate-Variants of Gated Recurrent Unit (GRU) Neural Networks

Gate-Variants of Gated Recurrent Unit (GRU) Neural Networks Gate-Variants of Gated Recurrent Unit (GRU) Neural Networks Rahul Dey and Fathi M. Salem Circuits, Systems, and Neural Networks (CSANN) LAB Department of Electrical and Computer Engineering Michigan State

More information

Deep Learning Basic Lecture - Complex Systems & Artificial Intelligence 2017/18 (VO) Asan Agibetov, PhD.

Deep Learning Basic Lecture - Complex Systems & Artificial Intelligence 2017/18 (VO) Asan Agibetov, PhD. Deep Learning 861.061 Basic Lecture - Complex Systems & Artificial Intelligence 2017/18 (VO) Asan Agibetov, PhD asan.agibetov@meduniwien.ac.at Medical University of Vienna Center for Medical Statistics,

More information

Fast Learning for Big Data Using Dynamic Function

Fast Learning for Big Data Using Dynamic Function IOP Conference Series: Materials Science and Engineering PAPER OPEN ACCESS Fast Learning for Big Data Using Dynamic Function To cite this article: T Alwajeeh et al 2017 IOP Conf. Ser.: Mater. Sci. Eng.

More information

Hidden Units. Sargur N. Srihari

Hidden Units. Sargur N. Srihari Hidden Units Sargur N. srihari@cedar.buffalo.edu 1 Topics in Deep Feedforward Networks Overview 1. Example: Learning XOR 2. Gradient-Based Learning 3. Hidden Units 4. Architecture Design 5. Backpropagation

More information

Keywords- Classification algorithm, Hypertensive, K Nearest Neighbor, Naive Bayesian, Data normalization

Keywords- Classification algorithm, Hypertensive, K Nearest Neighbor, Naive Bayesian, Data normalization GLOBAL JOURNAL OF ENGINEERING SCIENCE AND RESEARCHES APPLICATION OF CLASSIFICATION TECHNIQUES TO DETECT HYPERTENSIVE HEART DISEASE Tulasimala B. N* 1, Elakkiya S 2 & Keerthana N 3 *1 Assistant Professor,

More information

LSTM: An Image Classification Model Based on Fashion-MNIST Dataset

LSTM: An Image Classification Model Based on Fashion-MNIST Dataset LSTM: An Image Classification Model Based on Fashion-MNIST Dataset Kexin Zhang, Research School of Computer Science, Australian National University Kexin Zhang, U6342657@anu.edu.au Abstract. The application

More information

CHAPTER 6 DETECTION OF MASS USING NOVEL SEGMENTATION, GLCM AND NEURAL NETWORKS

CHAPTER 6 DETECTION OF MASS USING NOVEL SEGMENTATION, GLCM AND NEURAL NETWORKS 130 CHAPTER 6 DETECTION OF MASS USING NOVEL SEGMENTATION, GLCM AND NEURAL NETWORKS A mass is defined as a space-occupying lesion seen in more than one projection and it is described by its shapes and margin

More information

Facial Expression Classification with Random Filters Feature Extraction

Facial Expression Classification with Random Filters Feature Extraction Facial Expression Classification with Random Filters Feature Extraction Mengye Ren Facial Monkey mren@cs.toronto.edu Zhi Hao Luo It s Me lzh@cs.toronto.edu I. ABSTRACT In our work, we attempted to tackle

More information

Contents Machine Learning concepts 4 Learning Algorithm 4 Predictive Model (Model) 4 Model, Classification 4 Model, Regression 4 Representation

Contents Machine Learning concepts 4 Learning Algorithm 4 Predictive Model (Model) 4 Model, Classification 4 Model, Regression 4 Representation Contents Machine Learning concepts 4 Learning Algorithm 4 Predictive Model (Model) 4 Model, Classification 4 Model, Regression 4 Representation Learning 4 Supervised Learning 4 Unsupervised Learning 4

More information

Classification of Subject Motion for Improved Reconstruction of Dynamic Magnetic Resonance Imaging

Classification of Subject Motion for Improved Reconstruction of Dynamic Magnetic Resonance Imaging 1 CS 9 Final Project Classification of Subject Motion for Improved Reconstruction of Dynamic Magnetic Resonance Imaging Feiyu Chen Department of Electrical Engineering ABSTRACT Subject motion is a significant

More information

Global Optimality in Neural Network Training

Global Optimality in Neural Network Training Global Optimality in Neural Network Training Benjamin D. Haeffele and René Vidal Johns Hopkins University, Center for Imaging Science. Baltimore, USA Questions in Deep Learning Architecture Design Optimization

More information

Akarsh Pokkunuru EECS Department Contractive Auto-Encoders: Explicit Invariance During Feature Extraction

Akarsh Pokkunuru EECS Department Contractive Auto-Encoders: Explicit Invariance During Feature Extraction Akarsh Pokkunuru EECS Department 03-16-2017 Contractive Auto-Encoders: Explicit Invariance During Feature Extraction 1 AGENDA Introduction to Auto-encoders Types of Auto-encoders Analysis of different

More information

3D Densely Convolutional Networks for Volumetric Segmentation. Toan Duc Bui, Jitae Shin, and Taesup Moon

3D Densely Convolutional Networks for Volumetric Segmentation. Toan Duc Bui, Jitae Shin, and Taesup Moon 3D Densely Convolutional Networks for Volumetric Segmentation Toan Duc Bui, Jitae Shin, and Taesup Moon School of Electronic and Electrical Engineering, Sungkyunkwan University, Republic of Korea arxiv:1709.03199v2

More information

Domain-Aware Sentiment Classification with GRUs and CNNs

Domain-Aware Sentiment Classification with GRUs and CNNs Domain-Aware Sentiment Classification with GRUs and CNNs Guangyuan Piao 1(B) and John G. Breslin 2 1 Insight Centre for Data Analytics, Data Science Institute, National University of Ireland Galway, Galway,

More information

Fabric Image Retrieval Using Combined Feature Set and SVM

Fabric Image Retrieval Using Combined Feature Set and SVM Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology ISSN 2320 088X IMPACT FACTOR: 5.258 IJCSMC,

More information

SEMANTIC COMPUTING. Lecture 8: Introduction to Deep Learning. TU Dresden, 7 December Dagmar Gromann International Center For Computational Logic

SEMANTIC COMPUTING. Lecture 8: Introduction to Deep Learning. TU Dresden, 7 December Dagmar Gromann International Center For Computational Logic SEMANTIC COMPUTING Lecture 8: Introduction to Deep Learning Dagmar Gromann International Center For Computational Logic TU Dresden, 7 December 2018 Overview Introduction Deep Learning General Neural Networks

More information

Deep Learning for Computer Vision

Deep Learning for Computer Vision Deep Learning for Computer Vision Lecture 7: Universal Approximation Theorem, More Hidden Units, Multi-Class Classifiers, Softmax, and Regularization Peter Belhumeur Computer Science Columbia University

More information

Deep Learning for Computer Vision II

Deep Learning for Computer Vision II IIIT Hyderabad Deep Learning for Computer Vision II C. V. Jawahar Paradigm Shift Feature Extraction (SIFT, HoG, ) Part Models / Encoding Classifier Sparrow Feature Learning Classifier Sparrow L 1 L 2 L

More information

Lecture 13. Deep Belief Networks. Michael Picheny, Bhuvana Ramabhadran, Stanley F. Chen

Lecture 13. Deep Belief Networks. Michael Picheny, Bhuvana Ramabhadran, Stanley F. Chen Lecture 13 Deep Belief Networks Michael Picheny, Bhuvana Ramabhadran, Stanley F. Chen IBM T.J. Watson Research Center Yorktown Heights, New York, USA {picheny,bhuvana,stanchen}@us.ibm.com 12 December 2012

More information

Lecture 7: Neural network acoustic models in speech recognition

Lecture 7: Neural network acoustic models in speech recognition CS 224S / LINGUIST 285 Spoken Language Processing Andrew Maas Stanford University Spring 2017 Lecture 7: Neural network acoustic models in speech recognition Outline Hybrid acoustic modeling overview Basic

More information

Keywords:- Fingerprint Identification, Hong s Enhancement, Euclidian Distance, Artificial Neural Network, Segmentation, Enhancement.

Keywords:- Fingerprint Identification, Hong s Enhancement, Euclidian Distance, Artificial Neural Network, Segmentation, Enhancement. Volume 5, Issue 8, August 2015 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com Embedded Algorithm

More information

Cardiac Dysrhythmia Detection with GPU-Accelerated Neural Networks

Cardiac Dysrhythmia Detection with GPU-Accelerated Neural Networks Additional Features ECG Signal Cardiac Dysrhythmia Detection with GPU-Accelerated Neural Networks Albert Haque Computer Science Department, Stanford University AHAQUE@CS.STANFORD.EDU Abstract Cardiac dysrhythmia

More information

Hand Written Digit Recognition Using Tensorflow and Python

Hand Written Digit Recognition Using Tensorflow and Python Hand Written Digit Recognition Using Tensorflow and Python Shekhar Shiroor Department of Computer Science College of Engineering and Computer Science California State University-Sacramento Sacramento,

More information

arxiv: v1 [cs.cv] 20 Dec 2016

arxiv: v1 [cs.cv] 20 Dec 2016 End-to-End Pedestrian Collision Warning System based on a Convolutional Neural Network with Semantic Segmentation arxiv:1612.06558v1 [cs.cv] 20 Dec 2016 Heechul Jung heechul@dgist.ac.kr Min-Kook Choi mkchoi@dgist.ac.kr

More information

CAP 6412 Advanced Computer Vision

CAP 6412 Advanced Computer Vision CAP 6412 Advanced Computer Vision http://www.cs.ucf.edu/~bgong/cap6412.html Boqing Gong April 21st, 2016 Today Administrivia Free parameters in an approach, model, or algorithm? Egocentric videos by Aisha

More information

Face Recognition Using Vector Quantization Histogram and Support Vector Machine Classifier Rong-sheng LI, Fei-fei LEE *, Yan YAN and Qiu CHEN

Face Recognition Using Vector Quantization Histogram and Support Vector Machine Classifier Rong-sheng LI, Fei-fei LEE *, Yan YAN and Qiu CHEN 2016 International Conference on Artificial Intelligence: Techniques and Applications (AITA 2016) ISBN: 978-1-60595-389-2 Face Recognition Using Vector Quantization Histogram and Support Vector Machine

More information

Stochastic Function Norm Regularization of DNNs

Stochastic Function Norm Regularization of DNNs Stochastic Function Norm Regularization of DNNs Amal Rannen Triki Dept. of Computational Science and Engineering Yonsei University Seoul, South Korea amal.rannen@yonsei.ac.kr Matthew B. Blaschko Center

More information

Machine Learning With Python. Bin Chen Nov. 7, 2017 Research Computing Center

Machine Learning With Python. Bin Chen Nov. 7, 2017 Research Computing Center Machine Learning With Python Bin Chen Nov. 7, 2017 Research Computing Center Outline Introduction to Machine Learning (ML) Introduction to Neural Network (NN) Introduction to Deep Learning NN Introduction

More information

A CONFIGURABLE LOW POWER MIXED SIGNAL FOR PORTABLE ECG MONITORING SYSTEM

A CONFIGURABLE LOW POWER MIXED SIGNAL FOR PORTABLE ECG MONITORING SYSTEM A CONFIGURABLE LOW POWER MIXED SIGNAL FOR PORTABLE ECG MONITORING SYSTEM BAYYA RAMESH 1, P.NAGESWARA RAO 2 1 Bayya Ramesh, student, Vignan institute of Technology &Science, Hyderabad, Telangana, India.

More information

Deep Learning Cook Book

Deep Learning Cook Book Deep Learning Cook Book Robert Haschke (CITEC) Overview Input Representation Output Layer + Cost Function Hidden Layer Units Initialization Regularization Input representation Choose an input representation

More information

Research on the New Image De-Noising Methodology Based on Neural Network and HMM-Hidden Markov Models

Research on the New Image De-Noising Methodology Based on Neural Network and HMM-Hidden Markov Models Research on the New Image De-Noising Methodology Based on Neural Network and HMM-Hidden Markov Models Wenzhun Huang 1, a and Xinxin Xie 1, b 1 School of Information Engineering, Xijing University, Xi an

More information

DATA MINING AND MACHINE LEARNING. Lecture 6: Data preprocessing and model selection Lecturer: Simone Scardapane

DATA MINING AND MACHINE LEARNING. Lecture 6: Data preprocessing and model selection Lecturer: Simone Scardapane DATA MINING AND MACHINE LEARNING Lecture 6: Data preprocessing and model selection Lecturer: Simone Scardapane Academic Year 2016/2017 Table of contents Data preprocessing Feature normalization Missing

More information

A granular resampling method based energy-efficient architecture for heartbeat classification in ECG

A granular resampling method based energy-efficient architecture for heartbeat classification in ECG This article has been accepted and published on J-STAGE in advance of copyediting. Content is final as presented. IEICE Electronics Express, Vol.*, No.*, 1 10 A granular resampling method based energy-efficient

More information

3D-CNN and SVM for Multi-Drug Resistance Detection

3D-CNN and SVM for Multi-Drug Resistance Detection 3D-CNN and SVM for Multi-Drug Resistance Detection Imane Allaouzi, Badr Benamrou, Mohamed Benamrou and Mohamed Ben Ahmed Abdelmalek Essaâdi University Faculty of Sciences and Techniques, Tangier, Morocco

More information

Using neural nets to recognize hand-written digits. Srikumar Ramalingam School of Computing University of Utah

Using neural nets to recognize hand-written digits. Srikumar Ramalingam School of Computing University of Utah Using neural nets to recognize hand-written digits Srikumar Ramalingam School of Computing University of Utah Reference Most of the slides are taken from the first chapter of the online book by Michael

More information

Neural Network and Deep Learning. Donglin Zeng, Department of Biostatistics, University of North Carolina

Neural Network and Deep Learning. Donglin Zeng, Department of Biostatistics, University of North Carolina Neural Network and Deep Learning Early history of deep learning Deep learning dates back to 1940s: known as cybernetics in the 1940s-60s, connectionism in the 1980s-90s, and under the current name starting

More information

COMP9444 Neural Networks and Deep Learning 5. Geometry of Hidden Units

COMP9444 Neural Networks and Deep Learning 5. Geometry of Hidden Units COMP9 8s Geometry of Hidden Units COMP9 Neural Networks and Deep Learning 5. Geometry of Hidden Units Outline Geometry of Hidden Unit Activations Limitations of -layer networks Alternative transfer functions

More information

Sentiment Classification of Food Reviews

Sentiment Classification of Food Reviews Sentiment Classification of Food Reviews Hua Feng Department of Electrical Engineering Stanford University Stanford, CA 94305 fengh15@stanford.edu Ruixi Lin Department of Electrical Engineering Stanford

More information

EP-CapsNet: Extending Capsule Network with Inception Module for Electrophoresis Binary Classification

EP-CapsNet: Extending Capsule Network with Inception Module for Electrophoresis Binary Classification EP-CapsNet: Extending Capsule Network with Inception Module for Electrophoresis Binary Classification Elizabeth Tobing, Ashraf Murtaza, Keejun Han, Mun Y. Yi Graduate School of Knowledge Service Engineering

More information

ESPCI ParisTech, Laboratoire d Électronique, Paris France AMPS LLC, New York, NY, USA Hopital Lariboisière, APHP, Paris 7 University, Paris France

ESPCI ParisTech, Laboratoire d Électronique, Paris France AMPS LLC, New York, NY, USA Hopital Lariboisière, APHP, Paris 7 University, Paris France Efficient modeling of ECG waves for morphology tracking Rémi Dubois, Pierre Roussel, Martino Vaglio, Fabrice Extramiana, Fabio Badilini, Pierre Maison Blanche, Gérard Dreyfus ESPCI ParisTech, Laboratoire

More information

Partitioning Data. IRDS: Evaluation, Debugging, and Diagnostics. Cross-Validation. Cross-Validation for parameter tuning

Partitioning Data. IRDS: Evaluation, Debugging, and Diagnostics. Cross-Validation. Cross-Validation for parameter tuning Partitioning Data IRDS: Evaluation, Debugging, and Diagnostics Charles Sutton University of Edinburgh Training Validation Test Training : Running learning algorithms Validation : Tuning parameters of learning

More information

The Un-normalized Graph p-laplacian based Semi-supervised Learning Method and Speech Recognition Problem

The Un-normalized Graph p-laplacian based Semi-supervised Learning Method and Speech Recognition Problem Int. J. Advance Soft Compu. Appl, Vol. 9, No. 1, March 2017 ISSN 2074-8523 The Un-normalized Graph p-laplacian based Semi-supervised Learning Method and Speech Recognition Problem Loc Tran 1 and Linh Tran

More information

Practical Tips for using Backpropagation

Practical Tips for using Backpropagation Practical Tips for using Backpropagation Keith L. Downing August 31, 2017 1 Introduction In practice, backpropagation is as much an art as a science. The user typically needs to try many combinations of

More information

Deep Tracking: Biologically Inspired Tracking with Deep Convolutional Networks

Deep Tracking: Biologically Inspired Tracking with Deep Convolutional Networks Deep Tracking: Biologically Inspired Tracking with Deep Convolutional Networks Si Chen The George Washington University sichen@gwmail.gwu.edu Meera Hahn Emory University mhahn7@emory.edu Mentor: Afshin

More information

Network Traffic Measurements and Analysis

Network Traffic Measurements and Analysis DEIB - Politecnico di Milano Fall, 2017 Sources Hastie, Tibshirani, Friedman: The Elements of Statistical Learning James, Witten, Hastie, Tibshirani: An Introduction to Statistical Learning Andrew Ng:

More information