4,287 Hits in 7.0 sec

Extracting deep neural network bottleneck features using low-rank matrix factorization

Yu Zhang, Ekapol Chuangsuwanich, James Glass
2014 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)  
In this paper, we investigate the use of deep neural networks (DNNs) to generate a stacked bottleneck (SBN) feature representation for low-resource speech recognition.  ...  We examine different SBN extraction architectures, and incorporate low-rank matrix factorization in the final weight layer.  ...  Fig. 2 . 2 Diagram of the stacked bottleneck neural network feature extraction framework. Table 1 . 1 Turkish BN WERs for CI vs CD label training.  ... 
doi:10.1109/icassp.2014.6853583 dblp:conf/icassp/ZhangCG14 fatcat:ja7fh53w7rfyrbrgh27ovlmefm

An Unsupervised Deep Learning System for Acoustic Scene Analysis

Mou Wang, Xiao-Lei Zhang, Susanto Rahardja
2020 Applied Sciences  
It first extracts a bottleneck feature from the original acoustic feature of audio clips by an auto-encoder network, and then employs spectral clustering to further reduce the noise and unrelated information  ...  in the bottleneck feature.  ...  or deep convolutional neural network [7, 8] .  ... 
doi:10.3390/app10062076 fatcat:fpuk7np2p5ds3jh5uolgfl7nba

Pedestrian Re-identification Monitoring System Based on Deep Convolutional Neural Network

Wenzheng Qu, Zhiming Xu, Bei Luo, Haihua Feng, Zhiping Wan
2020 IEEE Access  
To solve these difficulties, this paper proposed a deep learning model and designed a system based on a deep convolutional neural network for pedestrian re-identification.  ...  INDEX TERMS Deep convolution, monitoring system, neural network, pedestrian re-identification. 86162 This work is licensed under a Creative Commons Attribution 4.0 License.  ...  In this paper, a convolutional neural network was employed to extract the feature vector of each pedestrian image.  ... 
doi:10.1109/access.2020.2986394 fatcat:azhmgxjg6vffjdxkjt6r7unkhe

Latent Factor Analysis of Deep Bottleneck Features for Speaker Verification with Random Digit Strings

Ziqiang Shi, Huibin Lin, Liu Liu, Rujie Liu
2018 Interspeech 2018  
This work investigates how to combine methods from deep bottleneck features (DBF) and latent factor analysis (LFA) to result in a new state-of-the-art approach for such task.  ...  In order to provide a wider temporal context, a stacked DBF is extracted to replace the traditional MFCC feature in the derivation of the supervector representations and leads to a significant improvement  ...  Once training is complete, all the layers after the bottleneck layer are removed, and the rest of the neural network is used to extract low-dimensional representation of the input.  ... 
doi:10.21437/interspeech.2018-1422 dblp:conf/interspeech/ShiLLL18b fatcat:cdzwulzsoveuxigrm7sxfd3y6i

Personalized POI Recommendation Based on Subway Network Features and Users' Historical Behaviors

Danfeng Yan, Xuan Zhao, Zhengkai Guo
2018 Wireless Communications and Mobile Computing  
Specifically, the subway network features such as the number of passing stations, waiting time, and transfer times are extracted and a recurrent neural network model is employed to model user behaviors  ...  , and subway network features.  ...  Data Availability The authors use a dataset collected from a real APP named Green Travel. It contains a large number of users, restaurants, and ratings for restaurant.  ... 
doi:10.1155/2018/3698198 fatcat:lpptebh4c5ey3hkloawtnajzd4

A study of rank-constrained multilingual DNNS for low-resource ASR

Reza Sahraeian, Dirk Van Compernolle
2016 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)  
Multilingual Deep Neural Networks (DNNs) have been successfully used to exploit out-of-language data to improve under-resourced ASR.  ...  In this paper, we improve on a multilingual DNN by utilizing low-rank factorization (LRF) of weight matrices via Singular Value Decomposition (SVD) to sparsify a multilingual DNN.  ...  LOW-RANK FACTORIZATION The use of low-rank matrix factorization for DNN training is proposed in [18] and [19] to reduce computational and space complexity for monolingual DNNs.  ... 
doi:10.1109/icassp.2016.7472713 dblp:conf/icassp/SahraeianC16 fatcat:jzxweeszwffapfxj3saaowejja

PILAE: A Non-gradient Descent Learning Scheme for Deep Feedforward Neural Networks [article]

P. Guo, K. Wang, X. L. Zhou
2021 arXiv   pre-print
In this work, a non-gradient descent learning (NGDL) scheme was proposed for deep feedforward neural networks (DNN).  ...  The PILAE with low rank approximation is a NGDL algorithm, and the encoder weight matrix is set to be the low rank approximation of the pseudoinverse of the input matrix, while the decoder weight matrix  ...  Alternatively, we could use other pre-trained deep networks to extract features which are then used as the inputs to the PILAE based network. 2).  ... 
arXiv:1811.01545v3 fatcat:xrdhlm26uvar7h66smbocnxlly

Deep matrix factorizations [article]

Pierre De Handschutter, Nicolas Gillis, Xavier Siebert
2020 arXiv   pre-print
Recently, deep matrix factorization (deep MF) was introduced to deal with the extraction of several layers of features and has been shown to reach outstanding performances on unsupervised tasks.  ...  Deep MF was motivated by the success of deep learning, as it is conceptually close to some neural networks paradigms.  ...  We believe deep MF could be a particularly useful framework as it combines the ability to extract hierarchical features, as deep learning models, with a high interpretability power, as low-rank matrix  ... 
arXiv:2010.00380v2 fatcat:5d6zleu6w5gh7nmduv2zxu7ep4

IGCV3: Interleaved Low-Rank Group Convolutions for Efficient Deep Neural Networks [article]

Ke Sun, Mingjie Li, Dong Liu, Jingdong Wang
2018 arXiv   pre-print
In this paper, we are interested in building lightweight and efficient convolutional neural networks.  ...  low-rank kernels, to form a convolutional kernel.  ...  Introduction There have been imperative demands for portable and efficient deep convolutional neural networks with high accuracies in vision applications.  ... 
arXiv:1806.00178v2 fatcat:dotuh6qahzeq3avar2saefz3ma

Online Embedding Compression for Text Classification using Low Rank Matrix Factorization [article]

Anish Acharya, Rahul Goel, Angeliki Metallinou, Inderjit Dhillon
2018 arXiv   pre-print
We propose a compression method that leverages low rank matrix factorization during training,to compress the word embedding layer which represents the size bottleneck for most NLP models.  ...  Deep learning models have become state of the art for natural language processing (NLP) tasks, however deploying these models in production system poses significant memory constraints.  ...  The low rank matrix factorization operation is illustrated in Figure 1 , where a single neural network matrix (layer) is replaced by two low rank matrices (layers).  ... 
arXiv:1811.00641v1 fatcat:7ywish3ymjerhmoforabop6rym

Recommendation system using a deep learning and graph analysis approach [article]

Mahdi Kherad, Amir Jalaly Bidgoly
2021 arXiv   pre-print
To solve these problems, context information such as user communication network is usually used.  ...  In addition, we leverage deep Autoencoders to initialize users and items latent factors, and deep embedding method gathers users' latent factors from the user trust graph.  ...  There are two ways to apply Autoencoder to the recommender system: 1) use Autoencoder to learn low-dimension features in the bottleneck layer and 2) fill in the blank entry of the ratings matrix in the  ... 
arXiv:2004.08100v8 fatcat:olpgxe5u5zg3tphqbofgdfilmu

Factorized Deep Neural Network Adaptation for Automatic Scoring of L2 Speech in English Speaking Tests

Dean Luo, Chunxiao Zhang, Linzhong Xia, Lixin Wang
2018 Interspeech 2018  
In this study, we investigate the effects of deep neural network factorized adaptation techniques on L2 speech assessment in real speaking tests.  ...  Factorizing the speaker, environment and other acoustic factors is crucial in evaluating L2 speech to effectively reduce acoustic mismatch between train and test conditions.  ...  features with extracted bottleneck features to train a DNN acoustic model. 4) At test time, extract normal iVectors from test data and pass them to exiting classification networks for factorization.  ... 
doi:10.21437/interspeech.2018-2138 dblp:conf/interspeech/LuoZXW18 fatcat:iouefnxa2zflxhf4hny325lq3i

Real-time Vehicle Detection implementing Deep Convolutional Neural Network features Data Augmentation Technique

V Sowmya, Research Department of Computer Science, SDNBV College for Women, University of Madras, Chrompet, Chennai-600044, India., R Radha
2022 Indian Journal of Science and Technology  
Essentially, the base network of a pre-trained deep model, fine-tuned VGG-16 is transformed into Faster R-CNN.  ...  Methods: This study proposes Faster Region-based Convolutional Neural Network (R-CNN) technique for image-based vehicle detection with significant performance benefits.  ...  The Deep Neural Network achieves significant accuracy in Regions with CNN features. The deeper architecture of CNN can solve the complex feature involved in object detection.  ... 
doi:10.17485/ijst/v15i1.1908 fatcat:aze4dsonancwrctxpr5jno52zu

Integrating Gaussian mixtures into deep neural networks: Softmax layer with hidden variables

Zoltan Tuske, Muhammad Ali Tahir, Ralf Schluter, Hermann Ney
2015 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)  
This paper shows that GMM can be easily integrated into the deep neural network framework.  ...  In contrast to this, in the tandem approach neural network output is used as input features to improve classic Gaussian mixture model (GMM) based emission probability estimates.  ...  Table 2 shows the baseline hybrid systems with different degree of low-rank factorization of the last matrix.  ... 
doi:10.1109/icassp.2015.7178779 dblp:conf/icassp/TuskeTSN15 fatcat:3xaucdgtrbf6tir7ionicudxx4

A Novel Discriminative Feature Extraction for Acoustic Scene Classification Using RNN Based Source Separation

Seongkyu MUN, Suwon SHON, Wooil KIM, David K. HAN, Hanseok KO
2017 IEICE transactions on information and systems  
scene classification, transfer learning, recurrent neural network, bottleneck feature  ...  Due to such sound classes being difficult to distinguish even by human ears, the conventional deep learning based feature extraction methods, as used by most DCASE participating teams, are considered facing  ...  Fig. 4 4 Proposed feature extraction based acoustic scene classification framework Fig. 5 5 Examples of source separation results (spectrograms) , convolutional neural network and Deep Neural Network  ... 
doi:10.1587/transinf.2017edl8132 fatcat:loymlit6g5ezjkk6ijuh5mdgou
« Previous Showing results 1 — 15 out of 4,287 results