In this study, we proposed an approach to report the condition of the eardrum as "normal" or "abnormal" by ensembling two different deep learning architectures. In the first network (Network 1), we...
详细信息
ISBN:
(数字)9781510616400
ISBN:
(纸本)9781510616400
In this study, we proposed an approach to report the condition of the eardrum as "normal" or "abnormal" by ensembling two different deep learning architectures. In the first network (Network 1), we applied transfer learning to the Inception V3 network by using 409 labeled samples. As a second network (Network 2), we designed a convolutional neural network to take advantage of auto-encoders by using additional 673 unlabeled eardrum samples. The individual classification accuracies of the Network 1 and Network 2 were calculated as 84.4%(+/- 12.1%) and 82.6% (+/- 11.3%), respectively. Only 32% of the errors of the two networks were the same, making it possible to combine two approaches to achieve better classification accuracy. The proposed ensemble method allows us to achieve robust classification because it has high accuracy (84.4%) with the lowest standard deviation (+/- 10.3%).
The advance prediction of seizures before its onset has been a challenging task for scientists for a long time. It is still the epileptic patients' hope to find an effective way of preventing seizures to improve t...
详细信息
ISBN:
(纸本)9781538668054
The advance prediction of seizures before its onset has been a challenging task for scientists for a long time. It is still the epileptic patients' hope to find an effective way of preventing seizures to improve the quality of their lives. In this paper, using an innovative mixing of unsupervised and supervised deep learning techniques, we propose a novel epileptic seizure prediction system using electroencephalogram (EEG) recordings from the human brains. The proposed system is built upon classifying between the interictal and the preictal brain states. The proposed system uses two-dimensional deep convolutional autoencoder for learning the best discriminative spatial features from the multichannel unlabeled raw EEG recordings. A Bidirectional Long Short-Term Memory recurrent neural network is used for classification based on the temporal information. To help achieve faster learning and reliable convergence for our system, the transfer learning technique is used for initializing the weights for the patient-specific networks. Within, up to one hour of prediction window, our system achieved an average sensitivity of 94.6% and average low false prediction alarm rate of 0.04FP/h which makes it one of the most efficient among state-of-the-art methods.
The focus of this paper is to illustrate how computational image processing and machine learning can help address two of the challenges of histological image analysis, namely, the cellular heterogeneity, and the impre...
详细信息
ISBN:
(数字)9781510607262
ISBN:
(纸本)9781510607255;9781510607262
The focus of this paper is to illustrate how computational image processing and machine learning can help address two of the challenges of histological image analysis, namely, the cellular heterogeneity, and the imprecise labeling. We propose an unsupervised method of generating representative image signatures based on an autoencoder architecture which reduces the dependency on labels that tend to be imprecise and tedious to get. We have modified and enhanced the architecture to simultaneously produce representative image features as well as perform dictionary learning on these features to enable robust characterization of the cellular phenotypes. We integrate the extracted features in a disease grading framework, test it in prostate tissues immunostained for different protein visualization and show significant improvement in terms of grading accuracy compared to alternative supervised feature-extraction methods.
The computer technology has shown strides of progress on vision related tasks, thanks to the application of Deep Learning (DL) and the convolutional Neural Networks (CNNs). The intro- duction in 2012 of the AlexNet CN...
详细信息
The computer technology has shown strides of progress on vision related tasks, thanks to the application of Deep Learning (DL) and the convolutional Neural Networks (CNNs). The intro- duction in 2012 of the AlexNet CNN, has demonstrated the power of the CNNs in computer vision related tasks. Since then there has been an ongoing effort to improve the performance of the CNNs. Their performance is affected by the initial values of their weights, their architectural characteristics and the quality and quantity of the data that they trains on. In the past it was popular to use a weight initialization scheme, by performing unsupervised pre-training using Con- volutional autoencoders (CAEs). Nowadays there is a preference towards more complex network architectures that perform well without the need for weight pre-conditioning. The increased com- plexity comes at a cost of computing power and memory bandwidth. The performance of these new networks is still dependent on the size and the quality of the training dataset. In this thesis we focus our research on improving object localization on images when we have a limited number of samples to train on. We demonstrate a novel method of CNN weight initialization, by pre-training using a network inspired by the autoencoders (AEs) and the U-Net network. This method allows a localizer CNN to achieve better generalization. We formulate this task as representation learning using multi-view supervised pre-training. We are taking advantage of complementary views of visual information in the training data. We propose a framework that includes two networks, the "AUX" and the "Localizer". The "AUX" network performs the pre-training using a main and an auxiliary view. The "Localizer" network performs the training for localization, using just the main view. The main view is an image that contains the object we intend to localize, and the auxiliary view is an image that contains abstract information about that object. We are leveraging the aux
We explore unsupervised representation learning of radio communication signals in raw sampled time series representation. We demonstrate that we can learn modulation basis functions using convolutional autoencoders an...
详细信息
ISBN:
(纸本)9781467389174
We explore unsupervised representation learning of radio communication signals in raw sampled time series representation. We demonstrate that we can learn modulation basis functions using convolutional autoencoders and visually recognize their relationship to the analytic bases used in digital communications. We also propose and evaluate quantitative metrics for quality of encoding using domain relevant performance metrics.
We introduce learned attention models into the radio machine learning domain for the task of modulation recognition by leveraging spatial transformer networks and introducing new radio domain appropriate transformatio...
详细信息
ISBN:
(纸本)9781538639542
We introduce learned attention models into the radio machine learning domain for the task of modulation recognition by leveraging spatial transformer networks and introducing new radio domain appropriate transformations. This attention model allows the network to learn a localization network capable of synchronizing and normalizing a radio signal blindly with zero knowledge of the signal's structure based on optimization of the network for classification accuracy, sparse representation, and regularization. Using this architecture we are able to outperform our prior results in accuracy vs signal to noise ratio against an identical system without attention, however we believe such an attention model has implication far beyond the task of modulation recognition.
暂无评论