Searcharxiv⌕ Search

arXiv subjects

Yi He

Publications and source records attributed to Yi He.

122 records · Page 7Linked to original sources
arXiv2019

Semi-supervised Skin Detection by Network with Mutual Guidance

In this paper we present a new data-driven method for robust skin detection from a single human portrait image. Unlike previous methods, we incorporate human body as a weak semantic guidance into this task, considering acquiring large-scale of human labeled skin data is commonly expensive and time-consuming. To be specific, we propose a dual-task neural network for joint detection of skin and body via a semi-supervised learning strategy. The dual-task network contains a shared encoder but two decoders for skin and body separately. For each decoder, its output also serves as a guidance for its counterpart, making both decoders mutually guided. Extensive experiments were conducted to demonstrate the effectiveness of our network with mutual guidance, and experimental results show our network outperforms the state-of-the-art in skin detection.

cs.CV↗
arXiv2019

Construction of New Generalizations of Wynn's Epsilon and Rho Algorithm by Solving Finite Difference Equations in the Transformation Order

We construct new sequence transformations based on Wynn's epsilon and rho algorithms. The recursions of the new algorithms include the recursions of Wynn's epsilon and rho algorithm and of Osada's generalized rho algorithm as special cases. We demonstrate the performance of our algorithms numerically by applying them to some linearly and logarithmically convergent sequences as well as some divergent series.

math.NA↗
arXiv2018

On the Convergence of Learning-based Iterative Methods for Nonconvex Inverse Problems

Numerous tasks at the core of statistics, learning and vision areas are specific cases of ill-posed inverse problems. Recently, learning-based (e.g., deep) iterative methods have been empirically shown to be useful for these problems. Nevertheless, integrating learnable structures into iterations is still a laborious process, which can only be guided by intuitions or empirical insights. Moreover, there is a lack of rigorous analysis about the convergence behaviors of these reimplemented iterations, and thus the significance of such methods is a little bit vague. This paper moves beyond these limits and proposes Flexible Iterative Modularization Algorithm (FIMA), a generic and provable paradigm for nonconvex inverse problems. Our theoretical analysis reveals that FIMA allows us to generate globally convergent trajectories for learning-based iterative methods. Meanwhile, the devised scheduling policies on flexible modules should also be beneficial for classical numerical methods in the nonconvex scenario. Extensive experiments on real applications verify the superiority of FIMA.

cs.CV↗
arXiv2018

Learning Collaborative Generation Correction Modules for Blind Image Deblurring and Beyond

Blind image deblurring plays a very important role in many vision and multimedia applications. Most existing works tend to introduce complex priors to estimate the sharp image structures for blur kernel estimation. However, it has been verified that directly optimizing these models is challenging and easy to fall into degenerate solutions. Although several experience-based heuristic inference strategies, including trained networks and designed iterations, have been developed, it is still hard to obtain theoretically guaranteed accurate solutions. In this work, a collaborative learning framework is established to address the above issues. Specifically, we first design two modules, named Generator and Corrector, to extract the intrinsic image structures from the data-driven and knowledge-based perspectives, respectively. By introducing a collaborative methodology to cascade these modules, we can strictly prove the convergence of our image propagations to a deblurring-related optimal solution. As a nontrivial byproduct, we also apply the proposed method to address other related tasks, such as image interpolation and edge-preserved smoothing. Plenty of experiments demonstrate that our method can outperform the state-of-the-art approaches on both synthetic and real datasets.

cs.CV↗
arXiv2018

Toward Designing Convergent Deep Operator Splitting Methods for Task-specific Nonconvex Optimization

Operator splitting methods have been successfully used in computational sciences, statistics, learning and vision areas to reduce complex problems into a series of simpler subproblems. However, prevalent splitting schemes are mostly established only based on the mathematical properties of some general optimization models. So it is a laborious process and often requires many iterations of ideation and validation to obtain practical and task-specific optimal solutions, especially for nonconvex problems in real-world scenarios. To break through the above limits, we introduce a new algorithmic framework, called Learnable Bregman Splitting (LBS), to perform deep-architecture-based operator splitting for nonconvex optimization based on specific task model. Thanks to the data-dependent (i.e., learnable) nature, our LBS can not only speed up the convergence, but also avoid unwanted trivial solutions for real-world tasks. Though with inexact deep iterations, we can still establish the global convergence and estimate the asymptotic convergence rate of LBS only by enforcing some fairly loose assumptions. Extensive experiments on different applications (e.g., image completion and deblurring) verify our theoretical results and show the superiority of LBS against existing methods.

cs.CV↗
arXiv2017

Partial-skew-orthogonal polynomials and related integrable lattices with Pfaffian tau-functions

Skew-orthogonal polynomials (SOPs) arise in the study of the n-point distribution function for orthogonal and symplectic random matrix ensembles. Motivated by the average of characteristic polynomials of the Bures random matrix ensemble studied in [22], we propose the concept of partial-skew-orthogonal polynomials (PSOPs) as a modification of the SOPs, and then the PSOPs with a variety of special skew-symmetric kernels and weight functions are addressed. By considering appropriate deformations of the weight functions, we derive nine integrable lattices in different dimensions. As a consequence, the tau-functions for these systems are shown to be expressed in terms of Pfaffians and the wave vectors PSOPs. In fact, the tau-functions also admit the representations of multiple integrals. Among these integrable lattices, some of them are known, while the others are novel to the best of our knowledge. In particular, one integrable lattice is related to the partition function of the Bures random matrix ensemble. Besides, we derive a discrete integrable lattice, which can be used to compute certain vector Padé approximants. This yields the first example regarding the connection between integrable lattices and vector Padé approximants, for which Hietarinta, Joshi and Nijhoff pointed out that "This field remains largely to be explored." in the recent monograph [27, Section 4.4] .

math-ph↗
arXiv2017

Ground state solution of fractional Schrödinger equations with a general nonlinearity

In this paper, we study the following fractional Schrödinger equation: \[ \left\{\begin{gathered} {(- Δ)^s}u + mu = f(u){\text{in}}{\mathbb{R}^N}, \hfill u \in {H^s}({\mathbb{R}^N}),{\text{}}u > 0{\text{on}}{\mathbb{R}^N}, \hfill \\ \end{gathered} \right. \] where $m>0$, $N>2s$, ${(- Δ)^s}$, $s \in (0,1)$ is the fractional Laplacian. Using minimax arguments, we obtain a positive ground state solution under general conditions on $f$ which we believe to be almost optimal.

math.AP↗
arXiv2017

Frame Stacking and Retaining for Recurrent Neural Network Acoustic Model

Frame stacking is broadly applied in end-to-end neural network training like connectionist temporal classification (CTC), and it leads to more accurate models and faster decoding. However, it is not well-suited to conventional neural network based on context-dependent state acoustic model, if the decoder is unchanged. In this paper, we propose a novel frame retaining method which is applied in decoding. The system which combined frame retaining with frame stacking could reduces the time consumption of both training and decoding. Long short-term memory (LSTM) recurrent neural networks (RNNs) using it achieve almost linear training speedup and reduces relative 41\% real time factor (RTF). At the same time, recognition performance is no degradation or improves sightly on Shenma voice search dataset in Mandarin.

cs.CL↗
arXiv2017

Deep LSTM for Large Vocabulary Continuous Speech Recognition

Recurrent neural networks (RNNs), especially long short-term memory (LSTM) RNNs, are effective network for sequential task like speech recognition. Deeper LSTM models perform well on large vocabulary continuous speech recognition, because of their impressive learning ability. However, it is more difficult to train a deeper network. We introduce a training framework with layer-wise training and exponential moving average methods for deeper LSTM models. It is a competitive framework that LSTM models of more than 7 layers are successfully trained on Shenma voice search data in Mandarin and they outperform the deep LSTM models trained by conventional approach. Moreover, in order for online streaming speech recognition applications, the shallow model with low real time factor is distilled from the very deep model. The recognition accuracy have little loss in the distillation process. Therefore, the model trained with the proposed training framework reduces relative 14\% character error rate, compared to original model which has the similar real-time capability. Furthermore, the novel transfer learning strategy with segmental Minimum Bayes-Risk is also introduced in the framework. The strategy makes it possible that training with only a small part of dataset could outperform full dataset training from the beginning.

cs.CL↗
arXiv2017

Exponential Moving Average Model in Parallel Speech Recognition Training

As training data rapid growth, large-scale parallel training with multi-GPUs cluster is widely applied in the neural network model learning currently.We present a new approach that applies exponential moving average method in large-scale parallel training of neural network model. It is a non-interference strategy that the exponential moving average model is not broadcasted to distributed workers to update their local models after model synchronization in the training process, and it is implemented as the final model of the training system. Fully-connected feed-forward neural networks (DNNs) and deep unidirectional Long short-term memory (LSTM) recurrent neural networks (RNNs) are successfully trained with proposed method for large vocabulary continuous speech recognition on Shenma voice search data in Mandarin. The character error rate (CER) of Mandarin speech recognition further degrades than state-of-the-art approaches of parallel training.

cs.CL↗
arXiv2014

Standing waves for a class of Schrödinger-Poisson equations in ${\mathbb{R}^3}$ involving critical Sobolev exponents

We are concerned with the following Schrödinger-Poisson equation with critical nonlinearity: \[\left\{\begin{gathered} - {\varepsilon ^2}Δu + V(x)u + ψu = λ|u{|^{p - 2}}u + |u{|^4}u{\text{in}}{\mathbb{R}^3}, \hfill - {\varepsilon ^2}Δψ= {u^2}{\text{in}}{\mathbb{R}^3},{\text{}}u > 0,{\text{}}u \in {H^1}({\mathbb{R}^3}), \hfill \end{gathered} \right. \] where $\varepsilon > 0$ is a small positive parameter, $λ> 0$, $3 < p \le 4$. Under certain assumptions on the potential $V$, we construct a family of positive solutions ${u_\varepsilon} \in {H^1}({\mathbb{R}^3})$ which concentrates around a local minimum of $V$ as $\varepsilon \to 0$.

math.AP↗
arXiv2013

Concentrating Bound States for Kirchhoff type problems in ${\R^3}$ involving critical Sobolev exponents

We study the concentration and multiplicity of weak solutions to the Kirchhoff type equation with critical Sobolev growth \[\left\{\begin{gathered} - \Bigl({\varepsilon ^2}a + \varepsilon b\int_{\R^3} {{{\left| {\nabla u} \right|}^2}} \Bigr)Δu + V(z)u = f(u) + {u^5}{\text{in}}{\R^3}, \hfill u \in {H^1}({\R^3}),{\text{}}u > 0{\text{in}}{\R^3}, \hfill \\ \end{gathered} \right.\] where $\varepsilon $ is a small positive parameter and $a,b > 0$ are constants, $f \in {C^1}({\R^ +},\R)$ is subcritical, $V:{\R^3} \to \R$ is a locally Hölder continuous function. We first prove that for ${\varepsilon_0} > 0$ sufficiently small, the above problem has a weak solution ${u_\varepsilon}$ with exponential decay at infinity. Moreover, ${u_\varepsilon}$ concentrates around a local minimum point of $V$ in $Λ$ as $\varepsilon \to 0$. With minimax theorems and Ljusternik-Schnirelmann theory, we also obtain multiple solutions by employing the topology construct of the set where the potential $V\left(z \right)$ attains its minimum.

math.AP↗
arXiv2010

Multistep epsilon-algorithm, Shanks' transformation, and Lotka-Volterra system by Hirota's method

In this paper, we give a multistep extension of the epsilon-algorithm of Wynn, and we show that it implements a multistep extension of the Shanks' sequence transformation which is defined by ratios of determinants. Reciprocally, the quantities defined in this transformation can be recursively computed by the multistep epsilon-algorithm. The multistep epsilon-algorithm and the multistep Shanks' transformation are related to an extended discrete Lotka-Volterra system. These results are obtained by using the Hirota's bilinear method, a procedure quite useful in the solution of nonlinear partial differential and difference equations.

math.NA↗