Classification of Known and Unknown Environmental Sounds Based on Self-Organized Space Using a Recurrent Neural Network

Our goal is to develop a system to learn and classify environmental sounds for robots working in the real world. In the real world, two main restrictions pertain in learning. (i) Robots have to learn using only a small amount of data in a limited time because of hardware restrictions. (ii) The system has to adapt to unknown data since it is virtually impossible to collect samples of all environmental sounds. We used a neuro-dynamical model to build a prediction and classification system. This neuro-dynamical model can self-organize sound classes into parameters by learning samples. The sound classification space, constructed by these parameters, is structured for the sound generation dynamics and obtains clusters not only for known classes, but also unknown classes. The proposed system searches on the basis of the sound classification space for classifying. In the experiment, we evaluated the accuracy of classification for both known and unknown sound classes.     

A System for Robotic Heart Surgery that Learns to Tie Knots Using Recurrent Neural Networks

Tying suture knots is a time-consuming task performed frequently during minimally invasive surgery (MIS). Automating this task could greatly reduce total surgery time for patients. Current solutions to this problem replay manually programmed trajectories, but a more general and robust approach is to use supervised machine learning to smooth surgeon-given training trajectories and generalize from them. Since knot tying generally requires a controller with internal memory to distinguish between identical inputs that require different actions at different points along a trajectory, it would be impossible to teach the system using traditional feedforward neural nets or support vector machines. Instead we exploit more powerful, recurrent neural networks (RNNs) with adaptive internal states. Results obtained using long short-term memory RNNs trained by the recent Evolino algorithm show that this approach can significantly increase the efficiency of suture knot tying in MIS over preprogrammed control.     

基于空间变换密集卷积网络的图片敏感文字识别

互联网上含有大量多字体混合、形变、拉伸、左右结构字形、倾斜畸变等复杂场景下的敏感文字图片,在处理相关图片过程中存在特征提取难、识别率低的问题.本文提出基于空间变换网络与密集神经网络的方法对图片敏感文字进行特征提取与变换矫正,使用了深层双向GRU网络与CTC时域连接网络对序列特征信息进行标记预测,序列化处理文本的方式可较好地提升距离较宽文字与模糊文字信息的处理能力.实验结果表明,本模型在Caffe-OCR中文合成数据集和CTW数据集中分别实现了87.0%和90.3%识别准确率,平均识别时间达到了26.3 ms/图.     

改进卷积神经网络的文本主题识别算法研究

针对于传统方法中存在的文本特征表示能力差、模型主题识别准确率低等问题,提出一种融合SENet与卷积神经网络的文本主题识别方法.将每个词对应的Word2vec词向量与LDA主题向量进行融合,并依据词语对主题的贡献度完成文档加权向量化处理;构建SECNN主题识别模型,使用SENet对卷积层输出的特征图进行权值的重标定,依靠...     

基于循环神经网络的藏语语音识别声学模型

探索将循环神经网络和连接时序分类算法应用于藏语语音识别声学建模,实现端到端的模型训练。同时根据声学模型输入与输出的关系,通过在隐含层输出序列上引入时域卷积操作来对网络隐含层时域展开步数进行约简,从而有效提升模型的训练与解码效率。实验结果显示,与传统基于隐马尔可夫模型的声学建模方法相比,循环神经网络模型在藏语拉萨话音素识别任务上具有更好的识别性能,而引入时域卷积操作的循环神经网络声学模型在保持同等识别性能的情况下,拥有更高的训练和解码效率。     

Non-Frontal Facial Expression Recognition Using a Depth-Patch Based Deep Neural Network

The challenge of coping with non-frontal head poses during facial expression recognition results in considerable reduction of accuracy and robustness when capturing expressions that occur during natural communications. In this paper, we attempt to recognize facial expressions under poses with large rotation angles from 2D videos. A depth-patch based 4D expression representation model is proposed. It was reconstructed from 2D dynamic images for delineating continuous spatial changes and temporal context under non-frontal cases. Furthermore, we present an effective deep neural network classifier, which can accurately capture pose-variant expression features from the depth patches and recognize non-frontal expressions. Experimental results on the BU-4DFE database show that the proposed method achieves a high recognition accuracy of 86.87% for non-frontal facial expressions within a range of head rotation angle of up to 52°, outperforming existing methods. We also present a quantitative analysis of the components contributing to the performance gain through tests on the BU-4DFE and Multi-PIE datasets.     

基于参数迁移和卷积循环神经网络的语音情感识别

在语音情感识别研究中,已有基于深度学习的方法大多没有针对语音时频两域的特征进行建模,且存在网络模型训练时间长、识别准确性不高等问题。语谱图是语音信号转换后具有时频两域的特殊图像,为了充分提取语谱图时频两域的情感特征,提出了一种基于参数迁移和卷积循环神经网络的语音情感识别模型。该模型把语谱图作为网络的输入,引入AlexNet网络模型并迁移其预训练的卷积层权重参数,将卷积神经网络输出的特征图重构后输入LSTM（Long Short-Term Memory）网络进行训练。实验结果表明,所提方法加快了网络训练的速度,并提高了情感识别的准确率。     

基于循环神经网络的语音识别模型

近年来基于隐马尔可夫模型（HMM）的语音识别技术得到了很大发展。然而HMM模型有着一定的局限性,如何克服HMM的一阶假设和独立性假设带来的问题一直是研究讨论的热点,在语音识别中引入神经网络的方法是克服HMM局限性的一条途径。该文将循环神经网络应用于汉语语音识别,修改了原网络模型并提出了相应的训练方法,实验结果表明该模型具有良好的连续信号处理性能,与传统的HMM模型效果相当,新的训练策略能够在提高训练速度的同时,使得模型分类性能有明显提高。     

神经网络在人脸识别中的应用

将BP神经网络用于人脸识别,并建立了人脸识别模型,该识别模型包括图像压缩、图像抽样、输入矢量标准化、BP神经网络与竞争选择处理过程,具有简单,识别率较高的特点。     

EMG-Based Motion Discrimination Using a Novel Recurrent Neural Network

This paper presents a pattern discrimination method for electromyogram (EMG) signals for application in the field of prosthetic control. The method uses a novel recurrent neural network based on the hidden Markov model. This network includes recurrent connections, which enable modeling time series, such as EMG signals. Weight coefficients in the network can be learned using a well-known back-propagation through time algorithm. Pattern discrimination experiments were conducted to demonstrate the feasibility and performance of the proposed method. We were able to successfully discriminate forearm motions using the EMG signals, and achieved considerably high discrimination performance compared with other discrimination methods.     

神经网络在人脸识别中的应用

将BP神经网络用于人脸识别,并建立了人脸识别模型,该识别模型包括图像压缩、图像抽样、输入矢量标准化、BP神经网络与竞争选择处理过程,具有简单,识别率较高的特点.     

噪声有源控制的递归神经网络方法

使用Filter－X算法研究有源噪声控制问题,存在需要较高阶次的滤波器和当噪声出现非线性时控制效果不佳的缺陷。为此提出一种基于对角递归神经网络的非线性噪声有源自适应控制方法,并给出一种基于误差梯度下降的在线学习算法,同时证明了闭环控制系统在Lyapunov意义下的稳定性。数值仿真表明,基于对解递归神经网络的噪声有源自适应控制是一种非常有效的噪声控制方法。     

Two-Axial Piezoelectric Actuator Controller Using a Multi-Layer Artificial Neural Network Featuring Feedback Connection for Tactile Displays

Although some compensation method is required when using a piezoelectric actuator because of hysteresis, a sensor feedback method is not suitable for an actuator array. In this study, we design a controller using a neural network to apply it to a tactile display composed of two-axial miniature actuators. This paper describes the two-axial miniature actuator, which is composed of two bimorph piezoelectric elements and two small links connected by three joints. A control system for the two-axial miniature actuator is designed based on a multi-layered artificial neural network to compensate for the hysteresis of piezoelectric elements. The output neuron emits the time derivative of voltage, a two-bit signal expressing increment or decrement condition is generated by two input neurons, and two input neurons calculate current values of voltage and displacement, respectively. The neural network is outfitted with a feedback loop including an integral element to reduce the number of neurons. In the experiment, if the result of the left piezoelectric element is compared to that of the right element, the displacement amplitudes and the inclinations coincide on the right and left piezoelectric elements. Although precise hysteresis characteristics such as loop width are considerably different, the present neural system can follow the difference.     

基于CSLSTM网络的文本情感分类

文本情感分类是自然语言处理领域的研究热点,更是产品评价领域的重要任务.考虑到词向量与句向量之间的语义关系和用户信息、产品信息对文本情感分类的影响,提出余弦相似度LSTM网络. 该网络通过在不同语义层级中引入用户信息和产品信息的注意力机制,并根据词向量和句向量之间的相似度初始化词层级注意力矩阵中隐层节点的权重. 在Yelp13、Yelp14和IMDB三个情感分类数据集上的实验结果表明文中方法的有效性.     

双向循环神经网络在语音识别中的应用

针对前馈神经网络难以处理时序数据的问题,提出将双向循环神经网络（BiRNN）应用在自动语音识别声学建模中。首先,应用梅尔频率倒谱系数进行特征提取;其次,采用双向循环神经网络作为声学模型;最后,测试不同参数对系统性能的影响。在TIMIT数据集上的实验结果表明,与基于卷积神经网络和深度神经网络的声学模型相比,识别率分别提升了1.3%和4.0%,说明基于双向循环神经网络的声学模型具有更好的性能。     

基于GDBN网络的文本情感倾向分类算法

情感倾向性分类是自然语言处理领域中的热门话题,它的一个重要应用是挖掘线上评论中的重要信息,掌握网络舆论走向,因此本文提出一种基于GDBN网络的文本情感倾向性分类算法.该算法通过引入遗传算法来改进深度置信网络模型中的隐层,使模型自行对隐单元个数寻优,取得当前模型的适宜值,并以此模型进行深层建模与特征提取.最后通过反向传播网络对提取到的特征进行情感倾向性分类.在多个文本数据集上进行实验验证,验证结果表明了本文算法的有效性.     

卷积神经网络的短文本分类方法

短文本分类是自然语言处理的一个研究热点.为提高文本分类精度和解决文本表示稀疏问题,提出了一种全新的文本表示（N-of-DOC）方法.采用Word2Vec分布式表示一个短语,将其转换成的向量作为卷积神经网络模型的输入,经过卷积层和池化层提取高层特征,输出层接分类器得出分类结果.实验结果表明,与传统机器学习（K近邻,支持向量机,逻辑斯特回归,朴素贝叶斯）相比,提出的方法不仅能解决中文文本向量的维数灾难和稀疏问题,而且在分类精度上也比传统方法提高了4.23%.     

Using a New Model of Recurrent Neural Network for Control

This paper shows the results obtained in controlling a mobile robot by means of local recurrent neural networks based on a radial basis function (RBF) type architecture. The model used has a Finite Impulse Response (FIR) filter feeding back each neuron's output to its own input, while using another FIR filter as a synaptic connection. The network parameters (coefficients of both filters) are adjusted by means of the gradient descent technique, thus obtaining the stability conditions of the process. As a practical application the system has been successfully used for controlling a wheelchair, using an architecture made up by a neurocontroller and a neuroidentifier. The role of the latter, connected up in parallel with the wheelchair, is to propagate the control error to the neurocontroller, thus cutting down the control error in each working cycle.     

基于小波神经网络的人脸识别

利用最近在小波变换、人工神经网络和证据理论上取得的进展来进行人脸图像的识别。由于小波变换在时间和频率空间具有良好的定位特性，使小波神经网络可对输入、输出数据进行多分辨的学习训练。将小波变换和反向传播神经网络理论结合，设计一种小波神经网络结构，介绍了神经网络的数学框架和该网络的学习算法，把此算法用到人脸识别中，实验结果证明小波神经网络在人脸识别中收敛速度快、识别率高。     

基于循环神经网络的语义完整性分析

本文提出了一种基于循环神经网络的语义完整性分析方法,通过判断句子是否语义完整,将长文本切分成多个语义完整句.首先,对文本进行分词,映射为相应的词向量并进行标注,然后将词向量和标注信息通过循环窗口和欠采样方法处理后,作为循环神经网络的输入,经过训练最后得到模型.实验结果表明,该方法可以达到91.61%的准确率,为主观题自动评分工作提供了基础,同时对语义分析、问答系统和机器翻译等研究有一定的帮助.