Search

Planning is crucial to train an agent via model-based reinforcement learning who can predict distant observations to reflect his/her past experience. Such a planning method is theoretically and computationally attractive in comparison with traditional learning which relies on step-by-step prediction. However, it is more challenging to build a learning machine which can predict and plan randomly across multiple time steps rather than act step by step. To reflect this flexibility in learning process, we need to predict future states directly without going through all intermediate states. Accordingly, this paper develops the stochastic temporal difference learning where the sequence data are represented with multiple jumpy states while the stochastic state space model is learned by maximizing the evidence lower bound of log likelihood of training data. A general solution with various number of jumpy states is developed and formulated. Experiments demonstrate the merit of the proposed sequential machine to find predictive states to roll forward with jumps as well as predict words.

Learning both domain mapping and domain knowledge is crucial for different sequence-to-sequence (seq2seq) tasks. Traditionally, seq2seq model only characterized domain mapping while the knowledge in source and target domains was ignored. To strengthen seq2seq representation, this study presents a unified transformer for bidirectional domain mapping where collaborative regularization is imposed. This regularization enforces the bidirectional mapping constraint and avoids the model from overfitting for better generalization. Importantly, the unified learning objective is optimized for collaborative learning among different modules in two domains with two learning directions. Experiments on machine translation demonstrate the merit of unified transformer by comparing with the existing methods under different tasks and settings.

Domain mapping aims to estimate a sophisticated mapping between source and target domains. Finding the specialized attribute in latent representation plays a key role to attain a desirable performance. However, the entangled features usually contain the mixed attribute which can not be easily decomposed in an unsupervised manner. To handle the mixed features for better generation, this paper presents an attribute decomposition based on the sequence data and carries out the flow-based image domain mapping. The latent variables, characterized by flow model, are decomposed into the attribute-relevant and attribute-irrelevant components. The decomposition is guided by multiple objectives including structural-perceptual loss, cycle consistency loss, sequential random-pair reconstruction loss and sequential classification loss where the paired training data for domain mapping are not required. Importantly, the sequential random-pair reconstruction loss is formulated by means of exchanging the attribute-relevant components within a sequence of images. As a result, the source images with the attributes of reference images can be smoothly transferred to the corresponding target images. Experiments on talking face synthesis show the merit of attribute decomposition in domain mapping.

Conditional variational autoencoder (cVAE) has shown promising performance in dialogue generation. However, there still exists two issues in dialog cVAE model. The first issue is the Kullback-Leiblier (KL) vanishing problem which results in degenerating cVAE into a simple recurrent neural network. The second issue is the assumption of isotropic Gaussian prior for latent variable which is too simple to assure diversity of the generated responses. To handle these issues, a simple distribution should be transformed into a complex distribution and simultaneously the value of KL divergence should be preserved. This paper presents the dialogue flow VAE (DF-VAE) for variational dialogue generation. In particular, KL vanishing is tackled by a new normalizing flow. An inverse autoregressive flow is proposed to transform isotropic Gaussian prior to a rich distribution. In the experiments, the proposed DF-VAE is significantly better than the other methods in terms of different evaluation metrics. The diversity of generated dialogue responses is enhanced. Ablation study is conducted to illustrate the merit of the proposed flow models.

Neural differential equation (NDE) is recently developed as a continuous-time state machine which can faithfully represent the irregularly-sampled sequence data. NDE is seen as a substantial extension of recurrent neural network (RNN) which conducts discrete-time state representation for regularly-sampled data. This study presents a new continuous-time attention to improve sequential learning where the region of interest in continuous-time state trajectory over observed as well as missing samples is sufficiently attended. However, the attention score, calculated by relating between a query and a sequence, is memory demanding because self-attention should treat all time observations as query vectors to feed them into ordinary differential equation (ODE) solver. To deal with this issue, we develop a new form of dynamics for continuous-time attention where the causality property is adopted such that query vector is fed into ODE solver up to current time. The experiments on irregularly-sampled human activities and medical features show that this method obtains desirable performance with efficient memory consumption.

This paper presents a new dual domain mapping based on a unified bidirectional sequence-to-sequence (seq2seq) learning. Traditionally, dual learning in domain mapping was constructed with intrinsic connection where the conditional generative models in two directions were mutually leveraged and combined. The additional feedback from the other generation direction was used to regularize sequential learning in original direction of domain mapping. Domain matching between source sequence and target sequence was accordingly improved. However, the reconstructions for knowledge in two domains were ignored. The dual information based on separate models in two training directions was not sufficiently discovered. To cope with this weakness, this study proposes a closed-loop seq2seq learning where domain mapping and domain knowledge are jointly learned. In particular, a new feature-level dual learning is incorporated to build a dualformer where feature integration and feature reconstruction are further performed to bridge dual tasks. Experiments demonstrate the merit of the proposed dualformer for machine translation based on the multi-objective seq2seq learning.

Latent variable model based on variational autoen-coder (VAE) is influential in machine learning for signal processing. VAE basically suffers from the issue of posterior collapse in sequential learning procedure where the variational posterior easily collapses to a prior as standard Gaussian. Latent semantics are then neglected in optimization process. The recurrent decoder therefore generates noninformative or repeated sequence data. To capture sufficient latent semantics from sequence data, this study simultaneously fulfills an amortized regularization for encoder, extends a Gaussian mixture prior for latent variable, and runs a skip connection for decoder. The noise robust prior, learned from the amortized encoder, is likely aware of temporal features. A variational prior based on the amortized mixture density is formulated in implementation of variational recurrent autoencoder for sequence reconstruction and representation. Owing to skip connection, the sequence samples are continuously predicted in decoder with contextual precision at each time step. Experiments on language model and sentiment classification show that the proposed method mitigates the issue of posterior collapse and learns the meaningful latent features to improve the inference and generation for semantic representation.

Model-free agent in reinforcement learning (RL) generally performs well but inefficient in training process with sparse data. A practical solution is to incorporate a model-based module in model-free agent. State transition can be learned to make desirable prediction of next state based on current state and action at each time step. This paper presents a new learning representation for variational RL by introducing the so-called transition uncertainty critic based on the variational encoder-decoder network where the uncertainty of structured state transition is encoded in a model-based agent. In particular, an action-gating mechanism is carried out to learn and decode the trajectory of actions and state transitions in latent variable space. The transition uncertainty maximizing exploration (TUME) is performed according to the entropy search by using the intrinsic reward based on the uncertainty measure corresponding to different states and actions. A dedicate latent variable model with a penalty using the bias of state-action value is developed. Experiments on Cart Pole and dialogue system show that the proposed TUME considerably performs better than the other exploration methods for reinforcement learning.

Learning across domains is challenging especially when test data in target domain are sparse, heterogeneous and unlabeled. This challenge is even severe when building a deep stochastic neural model. This paper presents a stochastic semi-supervised learning for domain adaptation by using labeled data from source domain and unlabeled data from target domain. There are twofold novelties in the proposed method. First, a graphical model is constructed to identify the random latent features for classes as well as domains which are learned by variational inference. Second, we learn the class features which are discriminative among classes and simultaneously invariant to both domains. An adversarial neural model is introduced to pursue domain invariance. The domain features are explicitly learned to purify the extraction of class features for an improved classification. The experiments on sentiment classification illustrate the merits of the proposed stochastic adversarial domain adaptation.

Recurrent neural network (RNN) has been widely used for sequential learning which has achieved a great success in different tasks. The temporal convolutional network (TCN), a variant of one-dimensional convolutional neural network (CNN), was also developed for sequential learning in presence of sequence data. RNN and TCN typically captures long-term and short-term features in temporal or spatial domain, respectively. This paper presents a new sequential learning, called the convolutional recurrent network (CRN), which fulfills TCN as an encoder and RNN as a decoder so that the global semantics as well as the local dependencies are simultaneously characterized from sequence data. To facilitate the interpretation and robustness in neural models, we further develop the stochastic modeling for CRN based on variational inference. The merits of CNN and RNN are then incorporated in inference of latent space which sufficiently produces a generative model for sequential prediction. Experiments on language model shows the effectiveness of stochastic CRN when compared with the other sequential machines.

Quantization on weight parameters in neural network training plays a key role for model compression in mobile devices. This paper presents a general M-ary adaptive quantization in construction of Bayesian neural networks. The trade-off between model capacity and memory cost is adjustable. The stochastic weight parameters are faithfully reflected. A compact model is trained to achieve robustness to model uncertainty due to heterogeneous data collection. To minimize the performance loss, the representation levels in quantized neural network are estimated by maximizing the variational lower bound of log likelihood conditioned on M-ary quantization. Bayesian learning is formulated by using the multi-spike-and- slab prior for quantization levels. An adaptive quantization is derived to implement a flexible parameter space for learning representation which is applied for object recognition. Experiments on image recognition show the merit of this Bayesian model compression for M-ary quantized neural networks.

Variational autoencoder (VAE) is a popular latent variable model for data generation. However, in natural language applications, VAE suffers from the posterior collapse in optimization procedure where the model posterior likely collapses to a standard Gaussian prior which disregards latent semantics from sequence data. The recurrent decoder accordingly generates du-plicate or noninformative sequence data. To tackle this issue, this paper adopts the Gaussian mixture prior for latent variable, and simultaneously fulfills the amortized regularization in encoder and skip connection in decoder. The noise robust prior, learned from the amortized encoder, becomes semantically meaningful. The prediction of sequence samples, due to skip connection, becomes contextually precise at each time. The amortized mixture prior (AMP) is then formulated in construction of variational recurrent autoencoder (VRAE) for sequence generation. Experiments on different tasks show that AMP-VRAE can avoid the posterior collapse, learn the meaningful latent features and improve the inference and generation for semantic representation.

Deep reinforcement learning (RL) is known as an emerging research trend in machine learning for autonomous systems. In real-world scenarios, the extrinsic rewards, acquired from the environment for learning an agent, are usually missing or extremely sparse. Such an issue of sparse reward constrains the learning capability of agent because the agent only updates the policy when the goal state is successfully attained. It is always challenging to implement an efficient exploration in RL algorithms. To tackle the sparse reward and inefficient exploration, the agent needs other helpful information to update its policy even when there is no interaction with the environment. This paper proposes the stochastic curiosity maximizing exploration (SCME), a learning strategy explored to allow the agent to act as human. We cope with the sparse reward problem by encouraging the agent to explore future diversity. To do so, a latent dynamic system is developed to acquire the latent states and latent actions to predict the variations in future conditions. The mutual information and the prediction error in the predicted states and actions are calculated as the intrinsic rewards. The agent based on SCME is therefore learned by maximizing these rewards to improve sample efficiency for exploration. The experiments on PyDial and Super Mario Bros show the benefits of the proposed SCME in dialogue system and computer game, respectively.

Parameter quantization is crucial for model compression. This paper generalizes the binary and ternary quantizations to M-ary quantization for adaptive learning of the quantized neural networks. To compensate the performance loss, the representation values and the quantization partitions of model parameters are jointly trained to optimize the resolution of gradients for parameter updating where the non-differentiable function in back-propagation algorithm is tackled. An asymmetric quantization is implemented. The restriction in parameter quantization is sufficiently relaxed. The resulting M-ary quantization scheme is general and adaptive with different M. Training of the M-ary quantized neural network (MQNN) can be tuned to balance the tradeoff between system performance and memory storage. Experimental results show that MQNN is able to achieve comparable image classification performance with full-precision neural network (FPNN), but the memory storage can be far less than that in FPNN.

Domain mismatch is a common problem in speaker verification. This paper proposes an information-maximized variational domain adversarial neural network (InfoVDANN) to reduce domain mismatch by incorporating an InfoVAE into domain adversarial training (DAT). DAT aims to produce speaker discriminative and domain-invariant features. The InfoVAE has two roles. First, it performs variational regularization on the learned features so that they follow a Gaussian distribution, which is essential for the standard PLDA backend. Second, it preserves mutual information between the features and the training set to extract extra speaker discriminative information. Experiments on both SRE16 and SRE18-CMN2 show that the InfoVDANN outperforms the recent VDANN, which suggests that increasing the mutual information between the latent features and input features enables the InfoVDANN to extract extra speaker information that is otherwise not possible.

Recurrent neural network (RNN) has achieved a great success in language modeling where the temporal information based on deterministic state is continuously extracted and evolved through time. Such a simple deterministic transition function using input-to-hidden and hidden-to-hidden weights is usually insufficient to reflect the diversities and variations of latent variable structure behind the heterogeneous natural language. This paper presents a new stochastic Markov RNN (MRNN) to strengthen the learning capability in language model where the trajectory of word sequences is driven by a neural Markov process with Markov state transitions based on a K-state long short-term memory model. A latent state machine is constructed to characterize the complicated semantics in the structured lexical patterns. Gumbel-softmax is introduced to implement the stochastic backpropatation algorithm with discrete states. The parallel computation for rapid realization of MRNN is presented. The variational Bayesian learning procedure is implemented. Experiments demonstrate the merits of stochastic and diverse representation using MRNN language model where the overhead of parameters and computations is limited.

This paper presents a new generative adversarial network (GAN) which artificially generates the i-vectors to compensate the imbalanced or insufficient data in speaker recognition based on the probabilistic linear discriminant analysis. Theoretically, GAN is powerful to generate the artificial data which are misclassified as the real data. However, GAN suffers from the mode collapse problem in two-player optimization over generator and discriminator. This study deals with this challenge by improving the model regularization through characterizing the weight uncertainty in GAN. A new Bayesian GAN is implemented to learn a regularized model from diverse data where the strong modes are flattened via the marginalization. In particular, we present a variational GAN (VGAN) where the encoder, generator and discriminator are jointly estimated according to the variational inference. The computation cost is significantly reduced. To assure the preservation of gradient values, the learning objective based on Wasserstein distance is further introduced. The issues of model collapse and gradient vanishing are alleviated. Experiments on NIST i-vector Speaker Recognition Challenge demonstrate the superiority of the proposed VGAN to the variational autoencoder, the standard GAN and the Bayesian GAN based on the sampling method. The learning efficiency and generation performance are evaluated.

Spatial image and optical flow provide complementary information for video representation and classification. Traditional methods separately encode two stream signals and then fuse them at the end of streams. This paper presents a new multi-stream recurrent neural network where streams are tightly coupled at each time step. Importantly, we propose a stochastic fusion mechanism for multiple streams of video data based on the Gumbel samples to increase the prediction power. A stochastic backpropagation algorithm is implemented to carry out a multi-stream neural network with stochastic fusion based on a joint optimization of convolutional encoder and recurrent decoder. Experiments on UCF101 dataset illustrate the merits of the proposed stochastic fusion in recurrent neural network in terms of interpretation and classification performance.

Generative adversarial network (GAN) has been successfully developing as a generative model where the artificial data drawn from the generator are misrecognized as real samples by a discriminator. Although GAN achieves the desirable performance, the challenge is that the mode collapse easily happens in the joint optimization of generator and discriminator. This study copes with this challenge by improving the model regularization by means of representing the weight uncertainty in GAN. A new Bayesian GAN is formulated and implemented to learn a regularized model from diverse data where the strong modes are flattened via the marginalization and the issues of model collapse and gradient vanishing are alleviated. In particular, we present a variational GAN (VGAN) where the encoder, generator and discriminator are jointly estimated according to the variational Bayesian inference. The experiments on image generation over two tasks (MNIST and CeleA) demonstrate the superiority of the proposed VGAN to the variational autoencoder, the standard GAN and the Bayesian GAN based on the sampling method. The learning efficiency and generation performance are evaluated.

An image classification system for a specific target domain is usually trained with initialization from a source domain given with a large number of classes, particularly in an application of image recognition. The classes in target domain are usually seen as a subset in source domain. Partial domain adaptation aims to tackle this generalization issue where no labeled data are provided in target domain. This paper presents an adversarial learning for partial domain adaptation where a symmetric metric based on the Wasserstein distance is adopted in an adversarial learning objective. We build a Wasserstein partial transfer network where the Wasserstein adversarial objective is jointly optimized to partially transfer the relevance knowledge from source to target domains. The geometric property for optimal transport is assured to mitigate the gradient vanishing problem in adversarial training. The neural network components for feature extraction, relevance transfer, domain matching and task classification are jointly trained by solving a minimax optimization over multiple objectives. Experiments on image classification show the merits of the proposed partially adversarial domain adaptation over different tasks.

Monaural source separation based on recurrent neural network is learned to characterize the sequential patterns in source signals based on dynamic states which are propagated through time. The hidden states are assumed to be deterministic along a single path where a shared long short-term memory (LSTM) is used. Such assumptions may not faithfully reflect the randomness and the variety of temporal features in mixed signals. To strengthen the capability of LSTM in source separation, we propose a stochastic Markov LSTM where the regression from the mixed signal to its source signals is learned with a stochastic indicator of Markov state which selects the state-dependent LSTM for signal separation at each time. A set of LSTMs is discovered to capture the structural diversity of temporal signals or the stochastic trajectory of state transitions for sequential prediction. A new state machine is constructed to learn the complicated latent semantics in heterogeneous and structural mappings between mixed signals and source signals. The Gumbel-softmax sampling is implemented in the backpropagation algorithm with discrete Markov states. Experiments on speech enhancement illustrate the merit of the proposed stochastic Markov LSTM in terms of short-term objective intelligibility measure of the separated speech.

How to overcome the training and test data mismatch in speaker verification systems has been a focus of research recently. In this paper, we propose a semi-supervised nuisance attribute network (SNAN) to reduce the domain mismatch in i-vectors and x-vectors. SNANs are based on the idea of nuisance attribute removal in inter-dataset variability compensation (IDVC). But instead of measuring the domain variability through the dataset means, SNANs use the maximum mean discrepancy (MMD) as part of their loss function, which enables the network to find nuisance directions in which domain variability is measured up to infinite moment. The architecture of SNANs also allows us to incorporate the out-of-domain speaker labels into the semi-supervised training process through the center loss and triplet loss. Using SNANs as a preprocessing step for PLDA training, we achieve a relative improvement of 11.8% in EER on NIST 2016 SRE compared to PLDA without adaptation. We also found that the semi-supervised approach can further improve SNANs’ performance.

Despite a great success in learning representation for image data, it is challenging to learn the stochastic latent features from natural language based on variational inference. The difficulty in stochastic sequential learning is due to the posterior collapse caused by an autoregressive decoder which is prone to be too strong to learn sufficient latent information during optimization. To compensate this weakness in learning procedure, a sophisticated latent structure is required to assure good convergence so that random features are sufficiently captured for sequential decoding. This study presents a new variational recurrent autoencoder (VRAE) for sequence reconstruction. There are two complementary encoders consisting of a long short-term memory (LSTM) and a pyramid bidirectional LSTM which are merged to discover the global and local dependencies in a hierarchical latent variable model, respectively. Experiments on Penn Treebank and Yelp 2013 demonstrate that the proposed hierarchical VRAE is able to learn the complementary representation as well as tackle the posterior collapse in stochastic sequential learning. The performance of recurrent autoencoder is substantially improved in terms of perplexity.

Long short-term memory (LSTM) has been successfully developed for monaural speech separation. Temporal information is learned by using dynamic states which are evolved through time and stored as an internal memory. The spectro-temporal data matrix of mixed signal is flattened as input vectors. There are twofold limitations. First, the internal memory in LSTM could not sufficiently characterize long-term information from different sources. Second, the temporal correlation and frequency neighboring in the flattened vectors were smeared. To deal with these limitations, this paper presents a convolutional neural Turing machine (ConvNTM) where the feature maps of spectro-temporal data are extracted and embedded in an external memory at each time step. ConvNTM aims to preserve the spectro-temporal structure in long sequential signals which is exploited to estimate the separated spectral signals. An addressing mechanism is introduced to continuously calculate the read and write heads to retrieve and update memory slots, respectively. The memory augmented source separation is implemented for single-channel speech enhancement. Experimental results illustrate the superiority of ConvNTM to LSTM, NTM and convolutional LSTM for speech enhancement in terms of short-term objective intelligibility measure.

Attention over natural language aims to spotlight on the meaningful region with representative keywords which can extract desirable meanings to accomplish a task of interest. The attention parameter is a latent variable which was indirectly estimated by minimizing the classification loss. For the task of question answering (QA), the classification loss may not sufficiently reflect the target answer. This paper proposes a direct solution which attends the meaningful region by minimizing the reconstruction loss due to auxiliary or supporting data which are available in different scenarios. In particular, minimizing the classification and reconstruction losses are carried out under the end-to-end memory network so that the memory-augmented question answering is realized. Such a supportive attention is implemented as a sequence-to-sequence model which reconstructs the supporting sentence to assure the translation invariance. The merit of this method is sufficiently demonstrated for sequential learning by using the bAbI QA and dialog tasks.

Deep learning has achieved great success in many real-world applications. For speech and language processing, recurrent neural networks are learned to characterize sequential patterns and extract the temporal information based on dynamic states which are evolved through time and stored as an internal memory. Traditionally, simple transition function using input-to-hidden and hidden-to-hidden weights is insufficient. To strengthen the learning capability, it is crucial to explore the diversity of latent structure in sequential signals and learn the stochastic trajectory of signal transitions to improve sequential prediction. This paper proposes the stochastic modeling of transitions in deep sequential learning. Our idea is to enhance latent variable representation by discovering the Markov state transitions in sequential data based on a K-state long short-term memory (LSTM) model. Such a latent state machine is capable of learning the complicated latent semantics in highly structured and heterogeneous sequential data. Gumbel-softmax is introduced to implement stochastic learning procedure with discrete states. Experimental results on visual and text language modeling illustrate the merit of the proposed stochastic transitions in sequential prediction with limited amount of parameters.

© 2018 IEEE. A Bayesian approach termed the BAyesian Least Squares Optimization with Nonnegative L1-norm constraint (BALSON) is proposed. The error distribution of data fitting is described by Gaussian likelihood. The parameter distribution is assumed to be a Dirichlet distribution. With the Bayes rule, searching for the optimal parameters is equivalent to finding the mode of the posterior distribution. In order to explicitly characterize the nonnegative L1-norm constraint of the parameters, we further approximate the true posterior distribution by a Dirichlet distribution. We estimate the moments of the approximated Dirichlet posterior distribution by sampling methods. Four sampling methods have been introduced and implemented. With the estimated posterior distributions, the original parameters can be effectively reconstructed in polynomial fitting problems, and the BALSON framework is found to perform better than conventional methods.

This study presents a spectro-temporal neural factorization (STNF) for speech dereverberation. Traditionally, a contextual window of spectro-temporal reverberant speech was unfolded into a one-way vector which was fed into a neural network to estimate the spectra of source speech at each time frame. Model parameters were trained by using the vectorized error backpropagation algorithm. System performance is constrained because contextual correlations and common factors in frequency and time horizons are disregarded. To compensate this weakness, a spectro-temporal factorization is incorporated to preserve the structural information in neural network training based on bi-factorized error backpropagation where the spectral and temporal factor matrices are estimated. Affine transformation in one-way neural network is generalized to the bilinear decomposition in bi-factorized neural network. The spectro-temporal features are extracted and forwarded to fully-connected layers for regression outputs. Such a STNF is further improved by merging with long short-term memory layer to capture the temporal features. Experiments results on 2014 REVERB Challenge demonstrate the meaningfulness of the factorized features and the merit of integrating these features for speech dereverberation.

This paper presents a novel memory-augmented neural network for single-channel source separation. We propose a recall neural network (RCNN) where a couple of external memories are realized for sequence-to-sequence learning based on an encoder and a decoder. These memories are learned in a two-pass sensing procedure where the mixed signal is encoded and then decoded (or recalled) as context vectors by using a bidirectional long short-term memory (LSTM) and a LSTM, respectively. These context vectors are integrated in a gating layer. A set of attention weights are calculated to attend the hidden state of decoder to implement a recurrent neural network for source separation. A gated attention mechanism is carried out to fulfill a specialized memory network. The regression errors due to two passes of sensing procedure and one pass of gated attention are jointly minimized to estimate the weight parameters of different components in different layers. The experiments on multi-speaker speech enhancement show that the proposed RCNN consistently outperforms LSTM and neural Turing machine in different settings.

This paper presents an adversarial manifold learning (AML) for speaker recognition based on the probabilistic linear discriminant analysis (PLDA) using i-vectors. PLDA basically consists of an encoder for finding the latent variables and a decoder for reconstructing the i-vectors. AML is developed and incorporated in deep learning for a latent variable model. Low-dimensional latent space is therefore constructed according to an adversarial learning with neighbor embedding. This AML-PLDA is formulated to jointly optimize three learning objectives including a reconstruction error based on PLDA, a subspace learning for neighbor embedding and an adversarial loss caused by a discriminator and a generator. Using the deep neural networks, the generator is trained to fool the discriminator with its generated samples in latent space. The parameters in encoder, decoder and discriminator are jointly estimated by using the stochastic gradient descent algorithm. The experiments on speaker recognition show the merit of AML-PLDA in manifold learning and pattern classification.

Traditional domain adaptation methods attempted to learn the shared representation for distribution matching between source domain and target domain where the individual information in both domains was not characterized. Such a solution suffers from the mixing problem of individual information with the shared features which considerably constrains the performance for domain adaptation. To relax this constraint, it is crucial to extract both shared information and individual information. This study captures both information via a new domain separation network where the shared features are extracted and purified via separate modeling of individual information in both domains. In particular, a hybrid adversarial learning is incorporated in a separation network as well as an adaptation network where the associated discriminators are jointly trained for domain separation and adaptation according to the minmax optimization over separation loss and domain discrepancy, respectively. Experiments on different tasks show the merit of using the proposed adversarial domain separation and adaptation.

Recurrent neural network (RNN) based on long short-term memory (LSTM) has been successfully developed for single-channel source separation. Temporal information is learned by using dynamic states which are evolved through time and stored as an internal memory. The performance of source separation is constrained due to the limitation of internal memory which could not sufficiently preserve long-term characteristics from different sources. This study deals with this limitation by incorporating an external memory in RNN and accordingly presents a memory augmented neural network for source separation. In particular, we carry out a neural Turing machine to learn a separation model for sequential signals of speech and noise in presence of different speakers and noise types. Experiments show that speech enhancement based on memory augmented neural network consistently outperforms that using deep neural network and LSTM in terms of short-term objective intelligibility measure.

This paper presents a variational manifold learning for speaker recognition based on the probabilistic linear discriminant analysis (PLDA) using i-vectors. A latent variable model is introduced to compensate the constraints of the linearity in PLDA scoring and the high dimensionality in using i-vectors. A deep variational learning is formulated to jointly optimize three objectives including a regularization for variational distributions, a reconstruction based on PLDA and a manifold learning for neighbor embedding. A stochastic gradient variational Bayesian algorithm is developed to optimize the variational lower bound of log likelihood where the expectation in the objectives is estimated via a sampling method. Interestingly, the latent variables in the proposed variational manifold PLDA (vm-PLDA) are capable of decoding or reconstructing the i-vectors. The experiments on visualization and speaker recognition show the merits of vm-PLDA in manifold learning and classification.

Speech recognition has been significantly improved by applying acoustic models based on deep neural network which could be realized as the feedforward NN (FNN) or the recurrent NN (RNN). In general, FNN is feasible to project the observations onto a deep invariant feature space while RNN is beneficial to capture the temporal information in a sequential data for speech recognition. RNN based on long short-term memory (LSTM) is capable of storing inputs over a long time period and thus exploiting a self-learned mechanism for long-range temporal context. Considering the complimentary FNN and RNN in their modeling capabilities, this paper presents a deep model which is constructed by stacking LSTM and FNN. Through the cascade of LSTM cells and fully-connected feedforward units, we explore the temporal patterns and summarize the long history of previous inputs in a deep learning machine. The experiments on 3rd CHiME challenge and Aurora-4 show that the stacks of hybrid model with FNN post-processor outperform stand-alone FNN and LSTM and the other hybrid models for robust speech recognition.

Deep neural network (DNN) is trained according to a mini-batch optimization based on the stochastic gradient descent algorithm. Such a stochastic learning suffers from instability in parameter updating and may easily trap into local optimum. This study deals with the stability of stochastic learning by reducing the variance of gradients in optimization procedure. We upgrade the optimization from the stochastic dual coordinated ascent (SDCA) to the accelerated SDCA without duality (or dual-free ASDCA). This optimization incorporates the momentum method to accelerate the updating rule where the variance of gradients can be reduced. Using dual-free ASDCA, the optimization of dual function of SDCA in a form of convex loss is implemented by directly optimizing the primal function with respect to pseudo-dual parameters. The non-convex optimization in DNN training can be resolved and accelerated. Experimental results illustrate the reduction of training loss, variance of gradients and word error rate by using the proposed optimization for DNN speech recognition.

This study presents a Bayesian approach to enhance the magnitude spectra of single-channel reverberant speech signals. Speech dereverberation model is constructed by using a nonnegative convolutive transfer function (NCTF) and a nonnegative matrix factorization (NMF). NCTF is used to characterize the magnitude spectra of speech signal and room impulse response while NMF is applied to represent the fine structure of speech spectra. Importantly, we deal with the variations of dereverberation model by introducing the exponential priors for reverberation kernel and noise signal. A full Bayesian solution to speech dereverberation is obtained according to the variational Bayesian inference algorithm. Using this algorithm, the room configuration and the speaker characteristics are automatically learned from data. Such a general model can be reduced to the previous methods. Experimental results on both simulated data and real recordings from 2014 REVERB Challenge show the merit of the proposed method for single-channel speech dereverberation.

Deep neural network is now a new trend towards solving different problems in speech processing. In this paper, we propose a discriminative deep recurrent neural network (DRNN) model for monaural speech separation. Our idea is to construct DRNN as a regression model to discover the deep structure and regularity for signal reconstruction from a mixture of two source spectra. To reinforce the discrimination capability between two separated spectra, we estimate DRNN separation parameters by minimizing an integrated objective function which consists of two measurements. One is the within-source reconstruction errors due to the individual source spectra while the other conveys the discrimination information which preserves the mutual difference between two source spectra during the supervised training procedure. This discrimination information acts as a kind of regularization so as to maintain between-source separation in monaural source separation. In the experiments, we demonstrate the effectiveness of the proposed method for speech separation compared with the other methods.