Pre-trained Deep Neural Network Research Articles

Task-specific feature extraction and classification of fMRI volumes using a deep neural network initialized with a deep belief network: Evaluation using sensorimotor tasks

Feedforward deep neural networks (DNNs), artificial neural networks with multiple hidden layers, have recently demonstrated a record-breaking performance in multiple areas of applications in computer vision and speech processing. Following the success, DNNs have been applied to neuroimaging modalities including functional/structural magnetic resonance imaging (MRI) and positron-emission tomography data. However, no study has explicitly applied DNNs to 3D whole-brain fMRI volumes and thereby extracted hidden volumetric representations of fMRI that are discriminative for a task performed as the fMRI volume was acquired. Our study applied fully connected feedforward DNN to fMRI volumes collected in four sensorimotor tasks (i.e., left-hand clenching, right-hand clenching, auditory attention, and visual stimulus) undertaken by 12 healthy participants. Using a leave-one-subject-out cross-validation scheme, a restricted Boltzmann machine-based deep belief network was pretrained and used to initialize weights of the DNN. The pretrained DNN was fine-tuned while systematically controlling weight-sparsity levels across hidden layers. Optimal weight-sparsity levels were determined from a minimum validation error rate of fMRI volume classification. Minimum error rates (mean±standard deviation; %) of 6.9 (±3.8) were obtained from the three-layer DNN with the sparsest condition of weights across the three hidden layers. These error rates were even lower than the error rates from the single-layer network (9.4±4.6) and the two-layer network (7.4±4.1). The estimated DNN weights showed spatial patterns that are remarkably task-specific, particularly in the higher layers. The output values of the third hidden layer represented distinct patterns/codes of the 3D whole-brain fMRI volume and encoded the information of the tasks as evaluated from representational similarity analysis. Our reported findings show the ability of the DNN to classify a single fMRI volume based on the extraction of hidden representations of fMRI volumes associated with tasks across multiple hidden layers. Our study may be beneficial to the automatic classification/diagnosis of neuropsychiatric and neurological diseases and prediction of disease severity and recovery in (pre-) clinical settings using fMRI volumes without requiring an estimation of activation patterns or ad hoc statistical evaluation.

Blind Estimation of Blur Kernels and Parameters from a Single Image

Image blur kernel estimation is critical to blind image deblurring. Most existing approaches exploit handcrafted blur features that are optimized for a certain uniform blur across the image, which is unrealistic in a real blind deconvolution setting, where the blur type is often unknown. To deal with this issue, we aim at identifying the blur type for each input image patch, and then estimating the kernel parameter in this paper. A learning-based method using a pre-trained deep neural network (DNN) and a general regression neural network (GRNN) is proposed to first classify the blur type and then estimate its parameters, taking advantages of both the classification ability of DNN and the regression ability of GRNN. To the best of our knowledge, this is the first time that pre-trained DNN and GRNN have been applied to the problem of blur analysis. First, our method identifies the blur type from a mixed input of image patches corrupted by various blurs with different parameters. To this aim, a supervised DNN is trained to project the input samples into a discriminative feature space, in which the blur type can be easily classified. Then, for each blur type, the proposed GRNN estimates the blur parameters with very high accuracy. Experiments demonstrate the effectiveness of the proposed method in several tasks with better or competitive results compared with the state of the art on two standard image data sets, i.e., the Berkeley segmentation data set and the Pascal VOC 2007 data set. In addition, blur region segmentation and deblurring on a number of real photographs show that our method outperforms the previous techniques even for non-uniformly blurred images.

Cough event classification by pretrained deep neural network.

BackgroundCough is an essential symptom in respiratory diseases. In the measurement of cough severity, an accurate and objective cough monitor is expected by respiratory disease society. This paper aims to introduce a better performed algorithm, pretrained deep neural network (DNN), to the cough classification problem, which is a key step in the cough monitor.MethodThe deep neural network models are built from two steps, pretrain and fine-tuning, followed by a Hidden Markov Model (HMM) decoder to capture tamporal information of the audio signals. By unsupervised pretraining a deep belief network, a good initialization for a deep neural network is learned. Then the fine-tuning step is a back propogation tuning the neural network so that it can predict the observation probability associated with each HMM states, where the HMM states are originally achieved by force-alignment with a Gaussian Mixture Model Hidden Markov Model (GMM-HMM) on the training samples. Three cough HMMs and one noncough HMM are employed to model coughs and noncoughs respectively. The final decision is made based on viterbi decoding algorihtm that generates the most likely HMM sequence for each sample. A sample is labeled as cough if a cough HMM is found in the sequence.ResultsThe experiments were conducted on a dataset that was collected from 22 patients with respiratory diseases. Patient dependent (PD) and patient independent (PI) experimental settings were used to evaluate the models. Five criteria, sensitivity, specificity, F1, macro average and micro average are shown to depict different aspects of the models. From overall evaluation criteria, the DNN based methods are superior to traditional GMM-HMM based method on F1 and micro average with maximal 14% and 11% error reduction in PD and 7% and 10% in PI, meanwhile keep similar performances on macro average. They also surpass GMM-HMM model on specificity with maximal 14% error reduction on both PD and PI.ConclusionsIn this paper, we tried pretrained deep neural network in cough classification problem. Our results showed that comparing with the conventional GMM-HMM framework, the HMM-DNN could get better overall performance on cough classification task.

High-order neural networks and kernel methods for peptide-MHC binding prediction.

Effective computational methods for peptide-protein binding prediction can greatly help clinical peptide vaccine search and design. However, previous computational methods fail to capture key nonlinear high-order dependencies between different amino acid positions. As a result, they often produce low-quality rankings of strong binding peptides. To solve this problem, we propose nonlinear high-order machine learning methods including high-order neural networks (HONNs) with possible deep extensions and high-order kernel support vector machines to predict major histocompatibility complex-peptide binding. The proposed high-order methods improve quality of binding predictions over other prediction methods. With the proposed methods, a significant gain of up to 25-40% is observed on the benchmark and reference peptide datasets and tasks. In addition, for the first time, our experiments show that pre-training with high-order semi-restricted Boltzmann machines significantly improves the performance of feed-forward HONNs. Moreover, our experiments show that the proposed shallow HONN outperform the popular pre-trained deep neural network on most tasks, which demonstrates the effectiveness of modelling high-order feature interactions for predicting major histocompatibility complex-peptide binding. There is no associated distributable software. renqiang@nec-labs.com or mark.gerstein@yale.edu Supplementary data are available at Bioinformatics online.

Dynamically pre-trained deep recurrent neural networks using environmental monitoring data for predicting PM2.5.

Fine particulate matter (hbox {PM}_{2.5}) has a considerable impact on human health, the environment and climate change. It is estimated that with better predictions, US$9 billion can be saved over a 10-year period in the USA (State of the science fact sheet air quality. http://www.noaa.gov/factsheets/new, 2012). Therefore, it is crucial to keep developing models and systems that can accurately predict the concentration of major air pollutants. In this paper, our target is to predict hbox {PM}_{2.5} concentration in Japan using environmental monitoring data obtained from physical sensors with improved accuracy over the currently employed prediction models. To do so, we propose a deep recurrent neural network (DRNN) that is enhanced with a novel pre-training method using auto-encoder especially designed for time series prediction. Additionally, sensors selection is performed within DRNN without harming the accuracy of the predictions by taking advantage of the sparsity found in the network. The numerical experiments show that DRNN with our proposed pre-training method is superior than when using a canonical and a state-of-the-art auto-encoder training method when applied to time series prediction. The experiments confirm that when compared against the hbox {PM}_{2.5} prediction system VENUS (National Institute for Environmental Studies. Visual Atmospheric Environment Utility System. http://envgis5.nies.go.jp/osenyosoku/, 2014), our technique improves the accuracy of hbox {PM}_{2.5} concentration level predictions that are being reported in Japan.

Fast Adaptation of Deep Neural Network Based on Discriminant Codes for Speech Recognition

Fast adaptation of deep neural networks (DNN) is an important research topic in deep learning. In this paper, we have proposed a general adaptation scheme for DNN based on discriminant condition codes, which are directly fed to various layers of a pre-trained DNN through a new set of connection weights. Moreover, we present several training methods to learn connection weights from training data as well as the corresponding adaptation methods to learn new condition code from adaptation data for each new test condition. In this work, the fast adaptation scheme is applied to supervised speaker adaptation in speech recognition based on either frame-level cross-entropy or sequence-level maximum mutual information training criterion. We have proposed three different ways to apply this adaptation scheme based on the so-called speaker codes: i) Nonlinear feature normalization in feature space; ii) Direct model adaptation of DNN based on speaker codes; iii) Joint speaker adaptive training with speaker codes. We have evaluated the proposed adaptation methods in two standard speech recognition tasks, namely TIMIT phone recognition and large vocabulary speech recognition in the Switchboard task. Experimental results have shown that all three methods are quite effective to adapt large DNN models using only a small amount of adaptation data. For example, the Switchboard results have shown that the proposed speaker-code-based adaptation methods may achieve up to 8-10% relative error reduction using only a few dozens of adaptation utterances per speaker. Finally, we have achieved very good performance in Switchboard (12.1% in WER) after speaker adaptation using sequence training criterion, which is very close to the best performance reported in this task (“Deep convolutional neural networks for LVCSR,” T. N. Sainath , Proc. IEEE Acoust., Speech, Signal Process., 2013).

Towards Scaling Up Classification-Based Speech Separation

Formulating speech separation as a binary classification problem has been shown to be effective. While good separation performance is achieved in matched test conditions using kernel support vector machines (SVMs), separation in unmatched conditions involving new speakers and environments remains a big challenge. A simple yet effective method to cope with the mismatch is to include many different acoustic conditions into the training set. However, large-scale training is almost intractable for kernel machines due to computational complexity. To enable training on relatively large datasets, we propose to learn more linearly separable and discriminative features from raw acoustic features and train linear SVMs, which are much easier and faster to train than kernel SVMs. For feature learning, we employ standard pre-trained deep neural networks (DNNs). The proposed DNN-SVM system is trained on a variety of acoustic conditions within a reasonable amount of time. Experiments on various test mixtures demonstrate good generalization to unseen speakers and background noises.

Speech Recognition Using Long-Span Temporal Patterns in a Deep Network Model

In recent years, there has been a renewed interest in the use of artificial neural networks (ANNs) for speech applications, and it seems that a new trend to move the speech technology forward has begun. Two main contributions have triggered such a new trend: 1) a major advance has been made in training the weights in deep neural networks (DNNs), and a pre-trained deep neural network hidden Markov model (DNN-HMM) hybrid architecture has outperformed a conventional Gaussian mixture model hidden Markov model (GMM-HMM) automatic speech recognition (ASR) system on a challenging business search dataset, and 2) it has been shown that phoneme classification can be boosted by using a hierarchical structure of multi-layer perceptrons (MLPs) trained to model long-span temporal patterns with beneficial effects on language recognition tasks. In this work, we combine these two lines of research and demonstrate that word recognition accuracy can be significantly enhanced by arranging DNNs in a hierarchical structure to model long-term energy trajectories. The proposed solution has been evaluated on the 5000-word Wall Street Journal task, resulting in consistent and significant improvements in both phone and word recognition accuracy rates. We have also analyzed the effects of various modeling choices on the system performance, and several architectural solutions have been compared.

Context-Dependent Pre-Trained Deep Neural Networks for Large-Vocabulary Speech Recognition

We propose a novel context-dependent (CD) model for large-vocabulary speech recognition (LVSR) that leverages recent advances in using deep belief networks for phone recognition. We describe a pre-trained deep neural network hidden Markov model (DNN-HMM) hybrid architecture that trains the DNN to produce a distribution over senones (tied triphone states) as its output. The deep belief network pre-training algorithm is a robust and often helpful way to initialize deep neural networks generatively that can aid in optimization and reduce generalization error. We illustrate the key components of our model, describe the procedure for applying CD-DNN-HMMs to LVSR, and analyze the effects of various modeling choices on performance. Experiments on a challenging business search dataset demonstrate that CD-DNN-HMMs can significantly outperform the conventional context-dependent Gaussian mixture model (GMM)-HMMs, with an absolute sentence accuracy improvement of 5.8% and 9.2% (or relative error reduction of 16.0% and 23.2%) over the CD-GMM-HMMs trained using the minimum phone error rate (MPE) and maximum-likelihood (ML) criteria, respectively.