Multiple Deep Neural Networks Research Articles

Deep Domain Generalization With Structured Low-Rank Constraint.

Domain adaptation nowadays attracts increasing interests in pattern recognition and computer vision field, since it is an appealing technique in fighting off weakly labeled or even totally unlabeled target data by leveraging knowledge from external well-learned sources. Conventional domain adaptation assumes that target data are still accessible in the training stage. However, we would always confront such cases in reality that the target data are totally blind in the training stage. This is extremely challenging since we have no prior knowledge of the target. In this paper, we develop a deep domain generalization framework with structured low-rank constraint to facilitate the unseen target domain evaluation by capturing consistent knowledge across multiple related source domains. Specifically, multiple domain-specific deep neural networks are built to capture the rich information within multiple sources. Meanwhile, a domain-invariant deep neural network is jointly designed to uncover most consistent and common knowledge across multiple sources so that we can generalize it to unseen target domains in the test stage. Moreover, structured low-rank constraint is exploited to align multiple domain-specific networks and the domain-invariant one in order to better transfer knowledge from multiple sources to boost the learning problem in unseen target domains. Extensive experiments are conducted on several cross-domain benchmarks and the experimental results show the superiority of our algorithm by comparing it with state-of-the-art domain generalization approaches.

Hierarchical Neural Representation of Dreamed Objects Revealed by Brain Decoding with Deep Neural Network Features.

Dreaming is generally thought to be generated by spontaneous brain activity during sleep with patterns common to waking experience. This view is supported by a recent study demonstrating that dreamed objects can be predicted from brain activity during sleep using statistical decoders trained with stimulus-induced brain activity. However, it remains unclear whether and how visual image features associated with dreamed objects are represented in the brain. In this study, we used a deep neural network (DNN) model for object recognition as a proxy for hierarchical visual feature representation, and DNN features for dreamed objects were analyzed with brain decoding of fMRI data collected during dreaming. The decoders were first trained with stimulus-induced brain activity labeled with the feature values of the stimulus image from multiple DNN layers. The decoders were then used to decode DNN features from the dream fMRI data, and the decoded features were compared with the averaged features of each object category calculated from a large-scale image database. We found that the feature values decoded from the dream fMRI data positively correlated with those associated with dreamed object categories at mid- to high-level DNN layers. Using the decoded features, the dreamed object category could be identified at above-chance levels by matching them to the averaged features for candidate categories. The results suggest that dreaming recruits hierarchical visual feature representations associated with objects, which may support phenomenal aspects of dream experience.

I-벡터 기반 오픈세트 언어 인식을 위한 다중 판별 DNN

본 논문에서는 여러 개의 이원 support vector machine (binary SVM)을 사용하여 세 개 이상의 클래스를 분류하는 multi-class SVM과 유사하게 다중의 판별 deep neural network (DNN) 모델을 사용하는 i-벡터 기반의 언어 인식 시스템을 제안한다. 제안하는 시스템은 NIST 2015 i-vector Machine Learning Challenge 데이터베이스에 포함된 i-벡터들을 이용하여 학습 및 테스트 되었으며, 오픈 세트에서 기존의 cosine distance, multi-class SVM 및 단일 neural network (NN) 기반의 언어 인식 시스템에 비하여 높은 성능을 보임이 확인되었다. In this paper, we propose an i-vector based language recognition system to identify the spoken language of the speaker, which uses multiple discriminative deep neural network (DNN) models analogous to the multi-class support vector machine (SVM) classification system. The proposed model was trained and tested using the i-vectors included in the NIST 2015 i-vector Machine Learning Challenge database, and shown to outperform the conventional language recognition methods such as cosine distance, SVM and softmax NN classifier in open-set experiments.

A Regression Approach to Single-Channel Speech Separation Via High-Resolution Deep Neural Networks

We propose a novel data-driven approach to single-channel speech separation based on deep neural networks (DNNs) to directly model the highly nonlinear relationship between speech features of a mixed signal containing a target speaker and other interfering speakers. We focus our discussion on a semisupervised mode to separate speech of the target speaker from an unknown interfering speaker, which is more flexible than the conventional supervised mode with known information of both the target and interfering speakers. Two key issues are investigated. First, we propose a DNN architecture with dual outputs of the features of both the target and interfering speakers, which is shown to achieve a better generalization capability than that with output features of only the target speaker. Second, we propose using a set of multiple DNNs, each intending to be signal-noise-dependent (SND), to cope with the difficulty that one single general DNN could not well accommodate all the speaker mixing variabilities at different signal-to-noise ratio (SNR) levels. Experimental results on the speech separation challenge (SSC) data demonstrate that our proposed framework achieves better separation results than other conventional approaches in a supervised or semisupervised mode. SND-DNNs could also yield significant performance improvements over a general DNN for speech separation in low SNR cases. Furthermore, for automatic speech recognition (ASR) following speech separation, this purely front-end processing with a single set of speaker-independent ASR acoustic models, achieves a relative word error rate (WER) reduction of 11.6% over a state-of-the-art separation and recognition system where a complicated joint back-end decoding framework with multiple sets of speaker-dependent ASR acoustic models needs to be implemented. When speaker-adaptive ASR acoustic models for the target speakers are adopted for the enhanced signals, another 12.1% WER reduction over our best speaker-independent ASR system is achieved.

Multichannel Audio Source Separation With Deep Neural Networks

This article addresses the problem of multichannel audio source separation. We propose a framework where deep neural networks (DNNs) are used to model the source spectra and combined with the classical multichannel Gaussian model to exploit the spatial information. The parameters are estimated in an iterative expectation-maximization (EM) fashion and used to derive a multichannel Wiener filter. We present an extensive experimental study to show the impact of different design choices on the performance of the proposed technique. We consider different cost functions for the training of DNNs, namely the probabilistically motivated Itakura–Saito divergence, and also Kullback–Leibler, Cauchy, mean squared error, and phase-sensitive cost functions. We also study the number of EM iterations and the use of multiple DNNs, where each DNN aims to improve the spectra estimated by the preceding EM iteration. Finally, we present its application to a speech enhancement problem. The experimental results show the benefit of the proposed multichannel approach over a single-channel DNN-based approach and the conventional multichannel nonnegative matrix factorization-based iterative EM algorithm.

Automatic Shadow Detection and Removal from a Single Image.

We present a framework to automatically detect and remove shadows in real world scenes from a single image. Previous works on shadow detection put a lot of effort in designing shadow variant and invariant hand-crafted features. In contrast, our framework automatically learns the most relevant features in a supervised manner using multiple convolutional deep neural networks (ConvNets). The features are learned at the super-pixel level and along the dominant boundaries in the image. The predicted posteriors based on the learned features are fed to a conditional random field model to generate smooth shadow masks. Using the detected shadow masks, we propose a Bayesian formulation to accurately extract shadow matte and subsequently remove shadows. The Bayesian formulation is based on a novel model which accurately models the shadow generation process in the umbra and penumbra regions. The model parameters are efficiently estimated using an iterative optimization procedure. Our proposed framework consistently performed better than the state-of-the-art on all major shadow databases collected under a variety of conditions.

A Deep Ensemble Learning Method for Monaural Speech Separation.

Monaural speech separation is a fundamental problem in robust speech processing. Recently, deep neural network (DNN)-based speech separation methods, which predict either clean speech or an ideal time-frequency mask, have demonstrated remarkable performance improvement. However, a single DNN with a given window length does not leverage contextual information sufficiently, and the differences between the two optimization objectives are not well understood. In this paper, we propose a deep ensemble method, named multicontext networks, to address monaural speech separation. The first multicontext network averages the outputs of multiple DNNs whose inputs employ different window lengths. The second multicontext network is a stack of multiple DNNs. Each DNN in a module of the stack takes the concatenation of original acoustic features and expansion of the soft output of the lower module as its input, and predicts the ratio mask of the target speaker; the DNNs in the same module employ different contexts. We have conducted extensive experiments with three speech corpora. The results demonstrate the effectiveness of the proposed method. We have also compared the two optimization objectives systematically and found that predicting the ideal time-frequency mask is more efficient in utilizing clean training speech, while predicting clean speech is less sensitive to SNR variations.

Ensemble of deep neural networks using acoustic environment classification for statistical model-based voice activity detection

In this paper, we investigate the ensemble of deep neural networks (DNNs) by using an acoustic environment classification (AEC) technique for the statistical model-based voice activity detection (VAD). From an investigation of the statistical model-based VAD, it is known that the traditional decision rule is based on the geometric mean of the likelihood ratio or the support vector machine (SVM), which is a shallow model with zero or one hidden layer. Since the shallow models cannot take an advantage of the diversity of the space distribution of features, in the training step, we basically build the multiple DNNs according the different noise types by employing the parameters of the statistical model-based VAD algorithm. In addition, the separate DNN is designed for the AEC algorithm in order to choose the best DNN for each noise. In the on-line noise-aware VAD step, the AEC is first performed on a frame-by-frame basis using the separate DNN so the a posteriori probabilities to identify noise are obtained. Once the probabilities are achieved for each noise, the environmental knowledge is contributed to allow us to combine the speech presence probabilities which are derived from the ensemble of the DNNs trained for the individual noise. Our approach for VAD was evaluated in terms of objective measures and showed significant improvement compared to the conventional algorithm.

Deep neural network based image annotation

Multilabel image annotation is one of the most important open problems in computer vision field. Unlike existing works that usually use conventional visual features to annotate images, features based on deep learning have shown potential to achieve outstanding performance. In this work, we propose a multimodal deep learning framework, which aims to optimally integrate multiple deep neural networks pretrained with convolutional neural networks. In particular, the proposed framework explores a unified two-stage learning scheme that consists of (i) learning to fine-tune the parameters of deep neural network with respect to each individual modality, and (ii) learning to find the optimal combination of diverse modalities simultaneously in a coherent process. Experiments conducted on a variety of public datasets evaluate the performance of the proposed framework for multilabel image annotation, in which the encouraging results validate the effectiveness of the proposed algorithms.

Multimodal deep network learning‐based image annotation

Multilabel image annotation is one of the most important open problems in the computer vision field. Unlike existing works that usually use conventional visual features to annotate images, features based on deep learning have shown potential to achieve outstanding performance. A multimodal deep learning framework is proposed, which aims to optimally integrate multiple deep neural networks pretrained with convolutional neural networks. In particular, the proposed framework explores a unified two-stage learning scheme that consists of (i) learning to fune-tune the parameters of the deep neural network with respect to each individual modality and (ii) learning to find the optimal combination of diverse modalities simultaneously in a coherent process. Experiments conducted on a variety of public datasets.

State-Clustering Based Multiple Deep Neural Networks Modeling Approach for Speech Recognition

The hybrid deep neural network (DNN) and hidden Markov model (HMM) has recently achieved dramatic performance gains in automatic speech recognition (ASR). The DNN-based acoustic model is very powerful but its learning process is extremely time-consuming. In this paper, we propose a novel DNN-based acoustic modeling framework for speech recognition, where the posterior probabilities of HMM states are computed from multiple DNNs (mDNN), instead of a single large DNN, for the purpose of parallel training towards faster turnaround. In the proposed mDNN method all tied HMM states are first grouped into several disjoint clusters based on data-driven methods. Next, several hierarchically structured DNNs are trained separately in parallel for these clusters using multiple computing units (e.g. GPUs). In decoding, the posterior probabilities of HMM states can be calculated by combining outputs from multiple DNNs. In this work, we have shown that the training procedure of the mDNN under popular criteria, including both frame-level cross-entropy and sequence-level discriminative training, can be parallelized efficiently to yield significant speedup. The training speedup is mainly attributed to the fact that multiple DNNs are parallelized over multiple GPUs and each DNN is smaller in size and trained by only a subset of training data. We have evaluated the proposed mDNN method on a 64-hour Mandarin transcription task and the 320-hour Switchboard task. Compared to the conventional DNN, a 4-cluster mDNN model with similar size can yield comparable recognition performance in Switchboard (only about 2% performance degradation) with a greater than 7 times speed improvement in CE training and a 2.9 times improvement in sequence training, when 4 GPUs are used.