DOCSLIB.ORG

When Are Tree Structures Necessary for Deep Learning of Representations?

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
When Are Tree Structures Necessary for Deep Learning of Representations? When Are Tree Structures Necessary for Deep Learning of Representations? Jiwei Li1, Minh-Thang Luong1, Dan Jurafsky1 and Eduard Hovy2 1Computer Science Department, Stanford University, Stanford, CA 94305 2Language Technology Institute, Carnegie Mellon University, Pittsburgh, PA 15213 jiweil,lmthang,[email protected] [email protected] Abstract For tasks where the inputs are larger text units (e.g., phrases, sentences or documents), a compo- Recursive neural models, which use syn- sitional model is first needed to aggregate tokens tactic parse trees to recursively generate into a vector with fixed dimensionality that can be representations bottom-up, are a popular used as a feature for other NLP tasks. Models for architecture. However there have not been achieving this usually fall into two categories: re- rigorous evaluations showing for exactly current models and recursive models: which tasks this syntax-based method is Recurrent models (also referred to as sequence appropriate. In this paper, we benchmark models) deal successfully with time-series data recursive neural models against sequential (Pearlmutter, 1989; Dorffner, 1996) like speech recurrent neural models, enforcing apples- (Robinson et al., 1996; Lippmann, 1989; Graves et to-apples comparison as much as possible. al., 2013) or handwriting recognition (Graves and We investigate 4 tasks: (1) sentiment clas- Schmidhuber, 2009; Graves, 2012). They were ap- sification at the sentence level and phrase plied early on to NLP (Elman, 1990), by modeling level; (2) matching questions to answer- a sentence as tokens processed sequentially and at phrases; (3) discourse parsing; (4) seman- each step combining the current token with pre- tic relation extraction. viously built embeddings. Recurrent models can Our goal is to understand better when, be extended to bidirectional ones from both left- and why, recursive models can outperform to-right and right-to-left. These models generally simpler models. We find that recursive consider no linguistic structure aside from word models help mainly on tasks (like seman- order. tic relation extraction) that require long- Recursive neural models (also referred to as tree distance connection modeling, particularly models), by contrast, are structured by syntactic on very long sequences. We then intro- parse trees. Instead of considering tokens sequen- duce a method for allowing recurrent mod- tially, recursive models combine neighbors based els to achieve similar performance: break- on the recursive structure of parse trees, starting ing long sentences into clause-like units from the leaves and proceeding recursively in a at punctuation and processing them sepa- bottom-up fashion until the root of the parse tree rately before combining. Our results thus is reached. For example, for the phrase the food help understand the limitations of both is delicious, following the operation sequence ( classes of models, and suggest directions (the food) (is delicious) ) rather than the sequen- for improving recurrent models. tial order (((the food) is) delicious). Many recur- sive models have been proposed (e.g., (Paulus et 1 Introduction al., 2014; Irsoy and Cardie, 2014)), and applied to various NLP tasks, among them entailment (Bow- Deep learning based methods learn low- man, 2013; Bowman et al., 2014), sentiment anal- dimensional, real-valued vectors for word ysis (Socher et al., 2013; Irsoy and Cardie, 2013; tokens, mostly from large-scale data corpus (e.g., Dong et al., 2014), question-answering (Iyyer et (Mikolov et al., 2013; Le and Mikolov, 2014; al., 2014), relation classification (Socher et al., Collobert et al., 2011)), successfully capturing 2012; Hashimoto et al., 2013), and discourse (Li syntactic and semantic aspects of text. and Hovy, 2014). 2304 Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2304–2314, Lisbon, Portugal, 17-21 September 2015. c 2015 Association for Computational Linguistics. One possible advantage of recursive models is ing is useful for finding similarities between their potential for capturing long-distance depen- question sentences and target phrases. dencies: two tokens may be structurally close to Semantic Relation Classification on the each other, even though they are far away in word • sequence. For example, a verb and its correspond- SemEval-2010 (Hendrickx et al., 2009) data ing direct object can be far away in terms of to- can help understand whether parsing is help- kens if many adjectives lies in between, but they ful in dealing with long-term dependencies, are adjacent in the parse tree (Irsoy and Cardie, such as relations between two words that are 2013). However we do not know if this advan- far apart in the sequence. tage is truly important, and if so for which tasks, Discourse parsing (RST dataset) is useful • or whether other issues are at play. Indeed, the for measuring the extent to which parsing im- reliance of recursive models on parsing is also a proves discourse tasks that need to combine potential disadvantage, given that parsing is rela- meanings of larger text units. Discourse pars- tively slow, domain-dependent, and can be error- ing treats elementary discourse units (EDUs) ful. as basic units to operate on, which are usually On the other hand, recent progress in multi- short clauses. The task also sheds light on ple subfields of neural NLP has suggested that re- the extent to which syntactic structures help current nets may be sufficient to deal with many acquire shot text representations. of the tasks for which recursive models have been proposed. Recurrent models without parse The principal motivation for this paper is to un- structures have shown good results in sequence- derstand better when, and why, recursive models to-sequence generation (Sutskever et al., 2014) are needed to outperform simpler models by en- for machine translation (e.g., (Kalchbrenner and forcing apples-to-apples comparison as much as Blunsom, 2013; 3; Luong et al., 2014)), pars- possible. This paper applies existing models to ing (Vinyals et al., 2014), and sentiment, where existing tasks, barely offering novel algorithms or for example recurrent-based paragraph vectors (Le tasks. Our goal is rather an analytic one, to inves- and Mikolov, 2014) outperform recursive models tigate different versions of recursive and recurrent (Socher et al., 2013) on the Stanford sentiment- models. This work helps understand the limita- bank dataset. tions of both classes of models, and suggest direc- Our goal in this paper is thus to investigate a tions for improving recurrent models. number of tasks with the goal of understanding The rest of this paper organized as follows: We for which kinds of problems recurrent models may detail versions of recursive/recurrent models in be sufficient, and for which kinds recursive mod- Section 2, present the tasks and results in Section els offer specific advantages. We investigate four 3, and conclude with discussions in Section 4. tasks with different properties. 2 Recursive and Recurrent Models Binary sentiment classification at the sen- 2.1 Notations • tence level (Pang et al., 2002) and phrase We assume that the text unit S, which could level (Socher et al., 2013) that focus on be a phrase, a sentence or a document, is com- understanding the role of recursive models prised of a sequence of tokens/words: S = in dealing with semantic compositionally in w , w , ..., w , where N denotes the num- { 1 2 NS } s various scenarios such as different lengths of ber of tokens in S. Each word w is associated inputs and whether or not supervision is com- with a K-dimensional vector embedding ew = prehensive. e1 , e2 , ..., eK . The goal of recursive and re- { w w w } current models is to map the sequence to a K- Phrase Matching on the UMD-QA dataset • dimensional eS, based on its tokens and their cor- (Iyyer et al., 2014) can help see the difference respondent embeddings. between outputs from intermediate compo- nents from different models, i.e., representa- Standard Recurrent/Sequence Models suc- tions for intermediate parse tree nodes and cessively take word wt at step t, combines its vec- outputs from recurrent models at different tor representation et with the previously built hid- time steps. It also helps see whether pars- den vector ht 1 from time t 1, calculates the re- − − 2305 sulting current embedding ht, and passes it to the next step. The embedding ht for the current time t is thus: it σ ft σ ht 1 = W − (5) ot σ · et ht = f(W ht 1 + V et) (1) · − · l tanh t where W and V denote compositional matrices. If ct = ft ct 1 + it lt (6) Ns denotes the length of the sequence, hNs repre- · − · sents the whole sequence S. hs = o c (7) t t · t 4K 2K Standard recursive/Tree models work in a where W R × . Labels at the ∈ similar way, but processing neighboring words by phrase/sentence level are predicted representations parse tree order rather than sequence order. It outputted from the last time step. computes a representation for each parent node based on its immediate children recursively in a Tree LSTMs Recent research has extended the bottom-up fashion until reaching the root of the LSTM idea to tree-based structures (Zhu et al., tree. For a given node η in the tree and its left child 2015; Tai et al., 2015) that associate memory and forget gates to nodes of the parse trees. ηleft (with representation eleft) and right child ηright (with representation eright), the standard recursive Bi-directional LSTMs These combine bi- network calculates eη as follows: directional models and LSTMs. e = f(W e + V e ) (2) η · ηleft · ηright 3 Experiments Bidirectional Models (Schuster and Paliwal, In this section, we detail our experimental settings 1997) add bidirectionality to the recurrent frame- and results. We consider the following tasks, each work where embeddings for each time are calcu- representative of a different class of NLP tasks.
Load more

Recommended publications
  • Transfer Learning Using CNN for Handwritten Devanagari Character
    Accepted for publication in IEEE International Conference on Advances in Information Technology (ICAIT), ICAIT - 2019, Please refer IEEE Explore for final version Transfer Learning using CNN for Handwritten Devanagari Character Recognition Nagender Aneja∗ and Sandhya Aneja ∗ ∗Universiti Brunei Darussalam Brunei Darussalam fnagender.aneja, sandhya.aneja [email protected] Abstract—This paper presents an analysis of pre-trained models to recognize handwritten Devanagari alphabets using transfer learning for Deep Convolution Neural Network (DCNN). This research implements AlexNet, DenseNet, Vgg, and Inception ConvNet as a fixed feature extractor. We implemented 15 epochs for each of AlexNet, DenseNet 121, DenseNet 201, Vgg 11, Vgg 16, Vgg 19, and Inception V3. Results show that Inception V3 performs better in terms of accuracy achieving 99% accuracy with average epoch time 16.3 minutes while AlexNet performs fastest with 2.2 minutes per epoch and achieving 98% accuracy. Index Terms—Deep Learning, CNN, Transfer Learning, Pre-trained, handwritten, recognition, Devanagari I. INTRODUCTION Handwriting identification is one of the challenging research domains in computer vision. Handwriting character identification systems are useful for bank signatures, postal code recognition, and bank cheques, etc. Many researchers working in this area tend to use the same database and compare different techniques that either perform better concerning the accuracy, time, complexity, etc. However, the handwritten recognition in non-English is difficult due to complex shapes. Fig. 1: Devanagari Alphabets Many Indian Languages like Hindi, Sanskrit, Nepali, Marathi, Sindhi, and Konkani uses Devanagari script. The alphabets of Devanagari Handwritten Character Dataset (DHCD) [1] is shown in Figure 1. Recognition of Devanagari is particularly challenging due to many groups of similar characters.
    [Show full text]
  • A Novel Approach to On-Line Handwriting Recognition Based on Bidirectional Long Short-Term Memory Networks
    A Novel Approach to On-Line Handwriting Recognition Based on Bidirectional Long Short-Term Memory Networks Marcus Liwicki1 Alex Graves2 Horst Bunke1 Jurgen¨ Schmidhuber2,3 1 Inst. of Computer Science and Applied Mathematics, University of Bern, Neubruckstr.¨ 10, 3012 Bern, Switzerland 2 IDSIA, Galleria 2, 6928 Manno-Lugano, Switzerland 3 TU Munich, Boltzmannstr. 3, 85748 Garching, Munich, Germany Abstract underlying this paper we aim at developing a handwriting recognition system to be used in a smart meeting room sce- In this paper we introduce a new connectionist approach nario [17], in our case the smart meeting room developed in to on-line handwriting recognition and address in partic- the IM2 project [11]. Smart meeting rooms usually have ular the problem of recognizing handwritten whiteboard multiple acquisition devices, such as microphones, cam- notes. The approach uses a bidirectional recurrent neu- eras, electronic tablets, and a whiteboard. In order to allow ral network with long short-term memory blocks. We use for indexing and browsing [18], automatic transcription of a recently introduced objective function, known as Connec- the recorded data is needed. tionist Temporal Classification (CTC), that directly trains In this paper, we introduce a novel approach to on-line the network to label unsegmented sequence data. Our new handwriting recognition, using a single recurrent neural net- system achieves a word recognition rate of 74.0 %, com- work (RNN) to transcribe the data. The key innovation is a pared with 65.4 % using a previously developed HMM- recently introduced RNN objective function known as Con- based recognition system. nectionist Temporal Classification (CTC) [5].
    [Show full text]
  • Unconstrained Online Handwriting Recognition with Recurrent Neural Networks
    Unconstrained Online Handwriting Recognition with Recurrent Neural Networks Alex Graves Santiago Fernandez´ Marcus Liwicki TUM, Germany IDSIA, Switzerland University of Bern, Switzerland [email protected] [email protected] [email protected] Horst Bunke Jurgen¨ Schmidhuber University of Bern, Switzerland IDSIA, Switzerland and TUM, Germany [email protected] [email protected] Abstract In online handwriting recognition the trajectory of the pen is recorded during writ- ing. Although the trajectory provides a compact and complete representation of the written output, it is hard to transcribe directly, because each letter is spread over many pen locations. Most recognition systems therefore employ sophisti- cated preprocessing techniques to put the inputs into a more localised form. How- ever these techniques require considerable human effort, and are specific to par- ticular languages and alphabets. This paper describes a system capable of directly transcribing raw online handwriting data. The system consists of an advanced re- current neural network with an output layer designed for sequence labelling, com- bined with a probabilistic language model. In experiments on an unconstrained online database, we record excellent results using either raw or preprocessed data, well outperforming a state-of-the-art HMM based system in both cases. 1 Introduction Handwriting recognition is traditionally divided into offline and online recognition. Offline recogni- tion is performed on images of handwritten text. In online handwriting the location of the pen-tip on a surface is recorded at regular intervals, and the task is to map from the sequence of pen positions to the sequence of words. At first sight, it would seem straightforward to label raw online inputs directly.
    [Show full text]
  • Start, Follow, Read: End-To-End Full Page Handwriting Recognition
    Start, Follow, Read: End-to-End Full-Page Handwriting Recognition Curtis Wigington1,2, Chris Tensmeyer1,2, Brian Davis1, William Barrett1, Brian Price2, and Scott Cohen2 1 Brigham Young University 2 Adobe Research [email protected] code at: https://github.com/cwig/start_follow_read Abstract. Despite decades of research, offline handwriting recognition (HWR) of degraded historical documents remains a challenging problem, which if solved could greatly improve the searchability of online cultural heritage archives. HWR models are often limited by the accuracy of the preceding steps of text detection and segmentation. Motivated by this, we present a deep learning model that jointly learns text detection, segmentation, and recognition using mostly images without detection or segmentation annotations. Our Start, Follow, Read (SFR) model is composed of a Region Proposal Network to find the start position of text lines, a novel line follower network that incrementally follows and preprocesses lines of (perhaps curved) text into dewarped images suitable for recognition by a CNN-LSTM network. SFR exceeds the performance of the winner of the ICDAR2017 handwriting recognition competition, even when not using the provided competition region annotations. Keywords: Handwriting Recognition, Document Analysis, Historical Document Processing, Text Detection, Text Line Segmentation. 1 Introduction In offline handwriting recognition (HWR), images of handwritten documents are converted into digital text. Though recognition accuracy on modern printed documents has reached acceptable performance for some languages [28], HWR for degraded historical documents remains a challenging problem due to large variations in handwriting appearance and various noise factors. Achieving ac- curate HWR in this domain would help promote and preserve cultural heritage by improving efforts to create publicly available transcriptions of historical doc- uments.
    [Show full text]
  • Contributions to Handwriting Recognition Using Deep Neural Networks and Quantum Computing Bogdan-Ionut Cirstea
    Contributions to handwriting recognition using deep neural networks and quantum computing Bogdan-Ionut Cirstea To cite this version: Bogdan-Ionut Cirstea. Contributions to handwriting recognition using deep neural networks and quantum computing. Artificial Intelligence [cs.AI]. Télécom ParisTech, 2018. English. tel-02412658 HAL Id: tel-02412658 https://hal.archives-ouvertes.fr/tel-02412658 Submitted on 15 Dec 2019 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. 2018-ENST-0059 EDITE - ED 130 Doctorat ParisTech THÈSE pour obtenir le grade de docteur délivré par TELECOM ParisTech Spécialité Signal et Images présentée et soutenue publiquement par Bogdan-Ionu¸tCÎRSTEA le 17 décembre 2018 Contributions à la reconnaissance de l’écriture manuscrite en utilisant des réseaux de neurones profonds et le calcul quantique Directrice de thèse : Laurence LIKFORMAN-SULEM T Jury Thierry PAQUET, Professeur, Université de Rouen Rapporteur H Christian VIARD-GAUDIN, Professeur, Université de Nantes Rapporteur È Nicole VINCENT, Professeur, Université Paris Descartes Examinateur Sandrine COURCINOUS, Expert, Direction Générale de l’Armement (DGA) Examinateur S Laurence LIKFORMAN-SULEM, Maître de conférence, HDR Télécom ParisTech Directrice de thèse E TELECOM ParisTech école de l’Institut Mines-Télécom - membre de ParisTech 46 rue Barrault 75013 Paris - (+33) 1 45 81 77 77 - www.telecom-paristech.fr Acknowledgments I would like to start by thanking my PhD advisor, Laurence Likforman-Sulem.
    [Show full text]
  • Unsupervised Feature Learning for Writer Identification
    Unsupervised Feature Learning for Writer Identification A Master’s Thesis Submitted to the Faculty of the Escola Tecnica` d’Enginyeria de Telecomunicacio´ de Barcelona Universitat Politecnica` de Catalunya by Advisors: Stefan Fiel Student: Manuel Keglevic Eduard Pallas` Arranz Robert Sablatnig and Josep R. Casas In partial fulfillment of the requirements for the degree of MASTER IN TELECOMMUNICATIONS ENGINEERING Barcelona, July 2018 Unsupervised Feature Learning for Writer Identification Eduard Pallas` Arranz Title of the thesis: Unsupervised feature learning for writer identification Author: Eduard Pallas` Arranz 1 Abstract Our work presents a research on unsupervised feature learning methods for writer identification and retrieval. We want to study the impact of deep learning alter- natives in this field by proposing methodologies which explore different uses of autoencoder networks. Taking a patch extraction algorithm as a starting point, we aim to obtain charac- teristics from patches of handwritten documents in an unsupervised way, mean- ing no label information is used for the task. To prove if the extraction of fea- tures is valid for writer identification, the approaches we propose are evaluated and compared with state-of-the-art methods on the ICDAR2013 and ICDAR2017 datasets for writer identification. Unsupervised Feature Learning for Writer Identification Eduard Pallas` Arranz Dedication: This thesis is dedicated to the loved ones waiting at home, to Giulia, and to all those who have made Vienna an unforgettable experience. 1 Unsupervised Feature Learning for Writer Identification Eduard Pallas` Arranz Acknowledgements First of all, I would like to sincerely thank Josep Ramon Casas for all the support and dedication of all these years, and also during the development of the thesis.
    [Show full text]
  • Fast Multi-Language LSTM-Based Online Handwriting Recognition
    Noname manuscript No. (will be inserted by the editor) Fast Multi-language LSTM-based Online Handwriting Recognition Victor Carbune · Pedro Gonnet · Thomas Deselaers · Henry A. Rowley · Alexander Daryin · Marcos Calvo · Li-Lun Wang · Daniel Keysers · Sandro Feuz · Philippe Gervais Received: Nov. 2018 / Accepted: - Abstract We describe an online handwriting system that is able to support 102 languages using a deep neu- ral network architecture. This new system has com- pletely replaced our previous Segment-and-Decode-based system and reduced the error rate by 20%-40% relative for most languages. Further, we report new state-of- the-art results on IAM-OnDB for both the open and closed dataset setting. The system combines methods from sequence recognition with a new input encoding using Bézier curves. This leads to up to 10x faster recog- nition times compared to our previous system. Through a series of experiments we determine the optimal con- figuration of our models and report the results of our Fig. 1 Example inputs for online handwriting recognition in dif- setup on a number of additional public datasets. ferent languages. See text for details. 1 Introduction alphabetic language, groups letters in syllables leading In this paper we discuss online handwriting recognition: to a large “alphabet” of syllables. Hindi writing often Given a user input in the form of an ink, i.e. a list of contains a connecting ‘Shirorekha’ line and characters touch or pen strokes, output the textual interpretation can form larger structures (grapheme clusters) which of this input. A stroke is a sequence of points (x; y; t) influence the written shape of the components.
    [Show full text]
  • A Connectionist Recognizer for On-Line Cursive Handwriting Recognition
    A CONNECTIONIST RECOGNIZER FOR ON-LINE CURSIVE HANDWRITING RECOGNITION Stefan Manke and Ulrich Bodenhausen University of Karlsruhe, Computer Science Department, D-76 128 Karlsruhe, Germany ABSTRACT Neural Network (MS-TU”), an extension of the TU”, coin- bines the high accuracy character recognition capahilities of ;I In this paper we show how the Multi-State Tim Delay Neural TDNN with a non-linear time alignment procedure (Dynamic Network (MS-TDNN), which is already used successfully in Time Warping) [7] for finding an optimal alignment between continuous speech recognition tasks, can be applied both to on- strokes and characters in handwritten continuous words. line single character and cursive (continuous) handwriting recog- The following section describes the basic network architecture nition. The MS-TDNN integrates the high accuracy single char- and training method of the MS-TDNN, followed by a description acter recognition capabilities of a TDNN with a non-linear time of the input features used in this paper (section 3). Section 4 pre- alignment procedure (dynamic time warping algorithm) for hd- sents results both on different writer dependent/writer indepn- ing stroke and character boundaries in isolated, handwritten char- dent, single character recognition task and writer dependent, acters ‘and words. In this approach each character is modelled by cursive handwriting recognition tasks. up to 3 different states and words are represented as a sequence of these characters. We describe the basic MS-TDNN architec- 2. THE MULTI-STATE TIME DELAY NEURAL ture and the input features used in this paper, and present results NETWORK (MS-TDNN) (up to 97.7% word recognition rate) both on writer dependent/ independent, single character recognition tasks and writer depen- The Time Delay Neural Network provides a time-shrft invariant dent, cursive handwriting tasks with varying vocahulary sizes up architecture for high performance on-line single character recog- to 20000 words.
    [Show full text]
  • Candidate Fusion: Integrating Language Modelling Into a Sequence-To-Sequence Handwritten Word Recognition Architecture
    Candidate Fusion: Integrating Language Modelling into a Sequence-to-Sequence Handwritten Word Recognition Architecture Lei Kang∗†, Pau Riba∗, Mauricio Villegasy, Alicia Forn´es∗, Mar¸calRusi~nol∗ ∗Computer Vision Center, Barcelona, Spain flkang, priba, afornes, [email protected] yomni:us, Berlin, Germany flei, [email protected] Abstract Sequence-to-sequence models have recently become very popular for tackling handwritten word recognition problems. However, how to effectively inte- grate an external language model into such recognizer is still a challenging problem. The main challenge faced when training a language model is to deal with the language model corpus which is usually different to the one used for training the handwritten word recognition system. Thus, the bias between both word corpora leads to incorrectness on the transcriptions, pro- viding similar or even worse performances on the recognition task. In this work, we introduce Candidate Fusion, a novel way to integrate an external language model to a sequence-to-sequence architecture. Moreover, it pro- vides suggestions from an external language knowledge, as a new input to the sequence-to-sequence recognizer. Hence, Candidate Fusion provides two improvements. On the one hand, the sequence-to-sequence recognizer has the flexibility not only to combine the information from itself and the lan- arXiv:1912.10308v1 [cs.CV] 21 Dec 2019 guage model, but also to choose the importance of the information provided by the language model. On the other hand, the external language model has the ability to adapt itself to the training corpus and even learn the most commonly errors produced from the recognizer.
    [Show full text]
  • Digit Recognition System Using Back Propagation Neural Network
    International Journal of Computer Science and Communication Vol. 2, No. 1, January-June 2011, pp. 197-205 DIGIT RECOGNITION SYSTEM USING BACK PROPAGATION NEURAL NETWORK Devinder Singh1 and Baljit Singh Khehra2 1Department of Computer Science, Mata Gujri College, Fatehgarh Sahib, INDIA, E-mail: [email protected] 2Department of Computer Science and Engineering, B.B.S.B. Engineering College, Fatehgarh Sahib, INDIA, E-mail: [email protected] ABSTRACT Digits are used in different types of data like vehicle number plate, numeric data, written data, meter reading etc. Digit recognition plays an important role in many user authentication applications in the modern world. In the proposed work, back propagation neural network based digit recognition system has been developed. Such system has four phases: preprocessing, segmentation, feature extraction and classification. When digit image is scanned, quality of image is degraded and some noise is added into this image. So it is necessary to reduce the noise and improve the quality of the digit image for OCR system. For this purpose, image enhancement is necessary. For this, frequency domain based Gausian filter is used to improve the quality and denoise the digit image. The goal of image segmentation is to separate the clear digit print area from the non-digit area. After segmentation, binary digit image is skeleton to reduce the width of digit into just a single line. In proposed system, coordinates bounding box, area, centroid, eccentricity, equiv-diameter features are used by back propagation neural network as a classifier to recognize digits. Experiments have been performed on standard dataset of digits.
    [Show full text]
  • Audio Word2vec: Unsupervised Learning of Audio Segment Representations Using Sequence-To-Sequence Autoencoder
    INTERSPEECH 2016 September 8–12, 2016, San Francisco, USA Audio Word2Vec: Unsupervised Learning of Audio Segment Representations using Sequence-to-sequence Autoencoder Yu-An Chung, Chao-Chung Wu, Chia-Hao Shen, Hung-Yi Lee, Lin-Shan Lee College of Electrical Engineering and Computer Science, National Taiwan University {b01902040, b01902038, r04921047, hungyilee}@ntu.edu.tw, [email protected] Abstract identification [4], and several approaches have been success- fully used in STD [10, 6, 7, 8]. But these vector representa- The vector representations of fixed dimensionality for tions may not be able to precisely describe the sequential pho- words (in text) offered by Word2Vec have been shown to be netic structures of the audio segments as we wish to have in this very useful in many application scenarios, in particular due paper, and these approaches were developed primarily in more to the semantic information they carry. This paper proposes a heuristic ways, rather than learned from data. Deep learning has parallel version, the Audio Word2Vec. It offers the vector rep- also been used for this purpose [12, 13]. By learning Recurrent resentations of fixed dimensionality for variable-length audio Neural Network (RNN) with an audio segment as the input and segments. These vector representations are shown to describe the corresponding word as the target, the outputs of the hidden the sequential phonetic structures of the audio segments to a layer at the last few time steps can be taken as the representation good degree, with very attractive real world applications such as of the input segment [13]. However, this approach is supervised query-by-example Spoken Term Detection (STD).
    [Show full text]
  • Handwriting Recognition Using LSTM Networks
    International Journal of New Technology and Research (IJNTR) ISSN:2454-4116, Volume-4, Issue-3, March 2018 Pages 116-119 Handwriting Recognition using LSTM Networks Sarita Yadav, Ankur Pandey, Pulkit Aggarwal, Rachit Garg, Vishal Aggarwal Abstract— Recognizing digits in an optimal way is a challenging problem. Recent deep learning based approaches have achieved great success on handwriting recognition. English characters are among the most widely adopted writing systems in the world. This paper presents a comparative evaluation of the standard LSTM RNN model with other deep models on MNIST dataset. Index Terms— LSTM, MNIST dataset. I. INTRODUCTION Recurrent neural networks (RNN) have recently shown great promise in tackling various sequence modeling tasks in machine learning, such as automatic speech recognition [1-2], language The paper presents a comparative evaluation of the standard LSTM translation [3-4], and generation of language descriptions for images RNN model with other deep models on MNIST dataset. [5-6]. Simple RNNs, however, are difficult to train using the stochastic gradient descent and have been reported to exhibit the II. THE RNN LSTM ARCHITECTURE so-called “vanishing” gradient and/or “exploding” gradient The RNN and LSTM are the class of Artificial Neural Network, phenomena [7-8]. This has limited the ability of simple RNN to which allow the network to preserve the dependency among learn sequences with relatively long dependencies. To address this sequence of input data. Unlink feedforward neural networks, RNNs limitation, researchers have developed a number of techniques in can use their internal state (memory) to process sequences of inputs. network architectures and optimization algorithms [9-11], among This makes them applicable to tasks such as unsegmented, which the most successful in applications is the Long Short-term connected handwriting recognition] or speech recognition.
    [Show full text]