DOCSLIB.ORG

Linear Prediction-Based Wavenet Speech Synthesis

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Linear Prediction-Based Wavenet Speech Synthesis LP-WaveNet: Linear Prediction-based WaveNet Speech Synthesis Min-Jae Hwang Frank Soong Eunwoo Song Search Solution Microsoft Naver Corporation Seongnam, South Korea Beijing, China Seongnam, South Korea [email protected] [email protected] [email protected] Xi Wang Hyeonjoo Kang Hong-Goo Kang Microsoft Yonsei University Yonsei University Beijing, China Seoul, South Korea Seoul, South Korea [email protected] [email protected] [email protected] Abstract—We propose a linear prediction (LP)-based wave- than the speech signal, the training and generation processes form generation method via WaveNet vocoding framework. A become more efficient. WaveNet-based neural vocoder has significantly improved the However, the synthesized speech is likely to be unnatural quality of parametric text-to-speech (TTS) systems. However, it is challenging to effectively train the neural vocoder when the target when the prediction errors in estimating the excitation are database contains massive amount of acoustical information propagated through the LP synthesis process. As the effect such as prosody, style or expressiveness. As a solution, the of LP synthesis is not considered in the training process, the approaches that only generate the vocal source component by synthesis output is vulnerable to the variation of LP synthesis a neural vocoder have been proposed. However, they tend to filter. generate synthetic noise because the vocal source component is independently handled without considering the entire speech To alleviate this problem, we propose an LP-WaveNet, production process; where it is inevitable to come up with a which enables to jointly train the complicated interactions mismatch between vocal source and vocal tract filter. To address between the excitation and LP synthesis filter. Based on the this problem, we propose an LP-WaveNet vocoder, where the basic assumption that the past speech samples and the LP complicated interactions between vocal source and vocal tract components are jointly trained within a mixture density network- coefficients are given as conditional information, we figure based WaveNet model. The experimental results verify that the out that the distributions of speech and excitation only lies on proposed system outperforms the conventional WaveNet vocoders a constant difference. Furthermore, if we model the speech both objectively and subjectively. In particular, the proposed distribution by using a mixture density network (MDN) [11], method achieves 4.47 MOS within the TTS framework. then the target speech distribution can be estimated by sum- Index Terms—Text-to-speech, speech synthesis, WaveNet ming the mean parameters of predicted mixture and an LP vocoder approximation, which is defined as the linear combination of I. INTRODUCTION past speech samples weighted by LP coefficients. Note that the LP-WaveNet is easy to train because the WaveNet only needs Waveform generation systems using WaveNet have signifi- to model the excitation component, and the complicated spec- cantly improved the synthesis quality of deep learning-based trum modeling part is embedded into the LP approximation. text-to-speech (TTS) systems [1]–[5]. Because the WaveNet vocoder can generate speech samples in a single unified In the objective and subjective evaluations, we verified arXiv:1811.11913v2 [eess.AS] 4 Mar 2020 neural network, it does not require any hand-engineered signal the outperforming performance of the proposed LP-WaveNet processing pipeline. Thus, it presents much higher synthetic in comparison to the conventional WaveNet-based neural quality than the traditional parametric vocoders [2]. vocoders. Especially, the LP-WaveNet provided 4.47 mean To further improve the perceptual quality of the synthesized opinion score result in the TTS framework. speech, more recent neural excitation vocoders take advantages of the merits from both the linear prediction (LP) vocoder II. WAVENET-BASED SPEECH SYNTHESIS SYSTEMS and the WaveNet structure [6]–[10]. In this framework, the A. µ-law quantization-based WaveNet formant-related spectral structure of the speech signal is decoupled by an LP analysis filter, and the WaveNet only WaveNet is a convolutional neural network (CNN)-based estimates the distribution of its residual signal (i.e., excitation). auto-regressive generative model that predicts the joint prob- Because the physical behavior of excitation signal is simpler ability distribution of speech samples x = fx1; x2; :::; xN g as follows: Work partially performed when the first author was an intern at Microsoft Y p(xjh) = p(xnjx<n; h); (1) Research Asia. n th where xn, x<n, and h denote the n speech sample, its III. LINEAR PREDICTION WAVENET VOCODER past speech samples, and the acoustic features, respectively. By stacking the dilated causal convolution layers multiply, the A. Fundamental mathematics WaveNet effectively extends its receptive field to the thousand Before introducing the proposed LP-WaveNet, a probabilis- of samples. tic relationship between speech and excitation signals have to The firstly proposed WaveNet, a.k.a., µ-law WaveNet [1], be clarified. Note that at the moment of nth sample generation defines the distribution of speech sample as a 256 categorical process in the WaveNet’s synthesis stage, x^n shown in Eq. (2) class of symbols obtained by an 8-bit µ-law quantized speech can be treated as a given factor since both LP coefficients, samples. To model the distribution of speech sample, the cate- ai, and previously reconstructed samples, fxn−ig are already gorical distribution is computed by applying softmax operation estimated. Hence, we conclude that the difference between two to the output of WaveNet. In the training phase, the weights random variables, xn and en, is only a known constant value of WaveNet is updated to minimize the cross-entropy loss. In term of x^n. the generation phase, the speech sample is auto-regressively Considering the shift property of second-order random generated in sample-by-sample. variable, if we define the speech’s distribution as a mixture of Since the µ-law WaveNet can generate the speech signal in Gaussian (MoG), the relationship between mixture parameters a single unified model, it provides significantly better synthetic of speech and excitation distributions can be lie on the only sound than the conventional parametric vocoders. However, it constant difference of mean parameters as follows: is not easy to train the network when the amount of database is larger and its acoustical informations such as prosody, style, or N " 2 # X wn;i (xn − µn;i) expressiveness are wider. Moreover, the synthesized sound of p(xnjx<n; h) = p exp − ; (3) 2πs 2s2 WaveNet is often suffered from the background noise artifact i=1 n;i n;i as the target speech signal is too coarsely quantized. x e wn;i =wn;i; B. WaveNet-based excitation modeling x e µn;i =µn;i + pn; (4) One effective solution is to model the excitation signal x e sn;i =sn;i; instead of the speech signal. For instance, in the ExcitNet approach [8], an excitation signal is first obtained by an LP where N and i denote the number and index of mixture, analysis filter, then its probabilistic behavior is trained by the respectively; w denotes the weights of mixture component; WaveNet framework. N (µ, s) imply the Gaussian distribution having mean of µ During the synthesis, the excitation signal is generated by and standard deviation of s; the superscripts e and x denote the trained WaveNet, then it is passed through an LP synthesis the excitation and the speech, respectively. Based on this filter to synthesize the speech signal as follows: observation, we propose an LP-WaveNet vocoder, where the LP synthesis process is structurally reflected to the WaveNet’s xn = en +x ^n; training and inference processes. p X (2) x^n = αixn−i; i=1 B. Network architecture th The detailed architecture of LP-WaveNet is illustrated in where en, x^n, p, and αi denote the n sample of excitation signal, the intermediate LP approximation term, the order of Fig. 1. In the proposed system, the distribution of speech sam- LP analysis, and the ith LP coefficient, respectively. Note that ple is defined as a MoG distribution by following Eq. (3), and the LP coefficients are periodically updated to match with the the LP-WaveNet is trained to generate the MoG parameters, extraction period of acoustic features. For instance, if acoustic [wn; µn; sn] conditioned by the input acoustic features. features are extracted at every 5-ms, then the LP coefficients In detail, the acoustic features pass through two 1- are updated at every 5-ms to synchronize the feature update dimensional convolution layers having kernel size of 3 for interval. explicitly imposing the contextual information of feature tra- Because the variation in the excitation signal is only con- jectory. Then, the residual connection with respect to the strained by vocal cord movement, its training is much easier input acoustic feature is applied to make the network more and the quality of finally synthesized speech is much higher, focus on the current frame information. Finally, the transposed too. However, the synthesized speech often contains unnatural convolution is applied to upsample the temporal resolution of artifacts because the excitation model is trained independently this features into that of speech signal. without considering the effect of LP synthesis filter; where To generate the speech samples, the mixture parameters, i.e., it happens mismatch between the excitation signal and LP mixture gain, mean and log-standard deviation, of excitation synthesis filter. To address this limitation, we propose an LP- signal are first predicted by WaveNet as follows: WaveNet, where both excitation signal and LP synthesis filter w µ s are jointly considered for training and synthesis. [zn ; zn; zn] = W aveNet(x<n; hn) (5) µ derivative of NLL loss L with respect to zn can be represented as follows: @L @L @µj µ = · µ : (8) @zi @µj @zi The time index n is omitted for the readability.
Load more

Recommended publications
  • Backpropagation and Deep Learning in the Brain
    Backpropagation and Deep Learning in the Brain Simons Institute -- Computational Theories of the Brain 2018 Timothy Lillicrap DeepMind, UCL With: Sergey Bartunov, Adam Santoro, Jordan Guerguiev, Blake Richards, Luke Marris, Daniel Cownden, Colin Akerman, Douglas Tweed, Geoffrey Hinton The “credit assignment” problem The solution in artificial networks: backprop Credit assignment by backprop works well in practice and shows up in virtually all of the state-of-the-art supervised, unsupervised, and reinforcement learning algorithms. Why Isn’t Backprop “Biologically Plausible”? Why Isn’t Backprop “Biologically Plausible”? Neuroscience Evidence for Backprop in the Brain? A spectrum of credit assignment algorithms: A spectrum of credit assignment algorithms: A spectrum of credit assignment algorithms: How to convince a neuroscientist that the cortex is learning via [something like] backprop - To convince a machine learning researcher, an appeal to variance in gradient estimates might be enough. - But this is rarely enough to convince a neuroscientist. - So what lines of argument help? How to convince a neuroscientist that the cortex is learning via [something like] backprop - What do I mean by “something like backprop”?: - That learning is achieved across multiple layers by sending information from neurons closer to the output back to “earlier” layers to help compute their synaptic updates. How to convince a neuroscientist that the cortex is learning via [something like] backprop 1. Feedback connections in cortex are ubiquitous and modify the
    [Show full text]
  • Training Autoencoders by Alternating Minimization
    Under review as a conference paper at ICLR 2018 TRAINING AUTOENCODERS BY ALTERNATING MINI- MIZATION Anonymous authors Paper under double-blind review ABSTRACT We present DANTE, a novel method for training neural networks, in particular autoencoders, using the alternating minimization principle. DANTE provides a distinct perspective in lieu of traditional gradient-based backpropagation techniques commonly used to train deep networks. It utilizes an adaptation of quasi-convex optimization techniques to cast autoencoder training as a bi-quasi-convex optimiza- tion problem. We show that for autoencoder configurations with both differentiable (e.g. sigmoid) and non-differentiable (e.g. ReLU) activation functions, we can perform the alternations very effectively. DANTE effortlessly extends to networks with multiple hidden layers and varying network configurations. In experiments on standard datasets, autoencoders trained using the proposed method were found to be very promising and competitive to traditional backpropagation techniques, both in terms of quality of solution, as well as training speed. 1 INTRODUCTION For much of the recent march of deep learning, gradient-based backpropagation methods, e.g. Stochastic Gradient Descent (SGD) and its variants, have been the mainstay of practitioners. The use of these methods, especially on vast amounts of data, has led to unprecedented progress in several areas of artificial intelligence. On one hand, the intense focus on these techniques has led to an intimate understanding of hardware requirements and code optimizations needed to execute these routines on large datasets in a scalable manner. Today, myriad off-the-shelf and highly optimized packages exist that can churn reasonably large datasets on GPU architectures with relatively mild human involvement and little bootstrap effort.
    [Show full text]
  • Q-Learning in Continuous State and Action Spaces
    -Learning in Continuous Q State and Action Spaces Chris Gaskett, David Wettergreen, and Alexander Zelinsky Robotic Systems Laboratory Department of Systems Engineering Research School of Information Sciences and Engineering The Australian National University Canberra, ACT 0200 Australia [cg dsw alex]@syseng.anu.edu.au j j Abstract. -learning can be used to learn a control policy that max- imises a scalarQ reward through interaction with the environment. - learning is commonly applied to problems with discrete states and ac-Q tions. We describe a method suitable for control tasks which require con- tinuous actions, in response to continuous states. The system consists of a neural network coupled with a novel interpolator. Simulation results are presented for a non-holonomic control task. Advantage Learning, a variation of -learning, is shown enhance learning speed and reliability for this task.Q 1 Introduction Reinforcement learning systems learn by trial-and-error which actions are most valuable in which situations (states) [1]. Feedback is provided in the form of a scalar reward signal which may be delayed. The reward signal is defined in relation to the task to be achieved; reward is given when the system is successfully achieving the task. The value is updated incrementally with experience and is defined as a discounted sum of expected future reward. The learning systems choice of actions in response to states is called its policy. Reinforcement learning lies between the extremes of supervised learning, where the policy is taught by an expert, and unsupervised learning, where no feedback is given and the task is to find structure in data.
    [Show full text]
  • Double Backpropagation for Training Autoencoders Against Adversarial Attack
    1 Double Backpropagation for Training Autoencoders against Adversarial Attack Chengjin Sun, Sizhe Chen, and Xiaolin Huang, Senior Member, IEEE Abstract—Deep learning, as widely known, is vulnerable to adversarial samples. This paper focuses on the adversarial attack on autoencoders. Safety of the autoencoders (AEs) is important because they are widely used as a compression scheme for data storage and transmission, however, the current autoencoders are easily attacked, i.e., one can slightly modify an input but has totally different codes. The vulnerability is rooted the sensitivity of the autoencoders and to enhance the robustness, we propose to adopt double backpropagation (DBP) to secure autoencoder such as VAE and DRAW. We restrict the gradient from the reconstruction image to the original one so that the autoencoder is not sensitive to trivial perturbation produced by the adversarial attack. After smoothing the gradient by DBP, we further smooth the label by Gaussian Mixture Model (GMM), aiming for accurate and robust classification. We demonstrate in MNIST, CelebA, SVHN that our method leads to a robust autoencoder resistant to attack and a robust classifier able for image transition and immune to adversarial attack if combined with GMM. Index Terms—double backpropagation, autoencoder, network robustness, GMM. F 1 INTRODUCTION N the past few years, deep neural networks have been feature [9], [10], [11], [12], [13], or network structure [3], [14], I greatly developed and successfully used in a vast of fields, [15]. such as pattern recognition, intelligent robots, automatic Adversarial attack and its defense are revolving around a control, medicine [1]. Despite the great success, researchers small ∆x and a big resulting difference between f(x + ∆x) have found the vulnerability of deep neural networks to and f(x).
    [Show full text]
  • Self-Training Wavenet for TTS in Low-Data Regimes
    StrawNet: Self-Training WaveNet for TTS in Low-Data Regimes Manish Sharma, Tom Kenter, Rob Clark Google UK fskmanish, tomkenter, [email protected] Abstract is increased. However, it can be seen from their results that the quality degrades when the number of recordings is further Recently, WaveNet has become a popular choice of neural net- decreased. work to synthesize speech audio. Autoregressive WaveNet is To reduce the voice artefacts observed in WaveNet stu- capable of producing high-fidelity audio, but is too slow for dent models trained under a low-data regime, we aim to lever- real-time synthesis. As a remedy, Parallel WaveNet was pro- age both the high-fidelity audio produced by an autoregressive posed, which can produce audio faster than real time through WaveNet, and the faster-than-real-time synthesis capability of distillation of an autoregressive teacher into a feedforward stu- a Parallel WaveNet. We propose a training paradigm, called dent network. A shortcoming of this approach, however, is that StrawNet, which stands for “Self-Training WaveNet”. The key a large amount of recorded speech data is required to produce contribution lies in using high-fidelity speech samples produced high-quality student models, and this data is not always avail- by an autoregressive WaveNet to self-train first a new autore- able. In this paper, we propose StrawNet: a self-training ap- gressive WaveNet and then a Parallel WaveNet model. We refer proach to train a Parallel WaveNet. Self-training is performed to models distilled this way as StrawNet student models. using the synthetic examples generated by the autoregressive We evaluate StrawNet by comparing it to a baseline WaveNet teacher.
    [Show full text]
  • Unsupervised Speech Representation Learning Using Wavenet Autoencoders Jan Chorowski, Ron J
    1 Unsupervised speech representation learning using WaveNet autoencoders Jan Chorowski, Ron J. Weiss, Samy Bengio, Aaron¨ van den Oord Abstract—We consider the task of unsupervised extraction speaker gender and identity, from phonetic content, properties of meaningful latent representations of speech by applying which are consistent with internal representations learned autoencoding neural networks to speech waveforms. The goal by speech recognizers [13], [14]. Such representations are is to learn a representation able to capture high level semantic content from the signal, e.g. phoneme identities, while being desired in several tasks, such as low resource automatic speech invariant to confounding low level details in the signal such as recognition (ASR), where only a small amount of labeled the underlying pitch contour or background noise. Since the training data is available. In such scenario, limited amounts learned representation is tuned to contain only phonetic content, of data may be sufficient to learn an acoustic model on the we resort to using a high capacity WaveNet decoder to infer representation discovered without supervision, but insufficient information discarded by the encoder from previous samples. Moreover, the behavior of autoencoder models depends on the to learn the acoustic model and a data representation in a fully kind of constraint that is applied to the latent representation. supervised manner [15], [16]. We compare three variants: a simple dimensionality reduction We focus on representations learned with autoencoders bottleneck, a Gaussian Variational Autoencoder (VAE), and a applied to raw waveforms and spectrogram features and discrete Vector Quantized VAE (VQ-VAE). We analyze the quality investigate the quality of learned representations on LibriSpeech of learned representations in terms of speaker independence, the ability to predict phonetic content, and the ability to accurately re- [17].
    [Show full text]
  • Unsupervised Speech Representation Learning Using Wavenet Autoencoders
    Unsupervised speech representation learning using WaveNet autoencoders https://arxiv.org/abs/1901.08810 Jan Chorowski University of Wrocław 06.06.2019 Deep Model = Hierarchy of Concepts Cat Dog … Moon Banana M. Zieler, “Visualizing and Understanding Convolutional Networks” Deep Learning history: 2006 2006: Stacked RBMs Hinton, Salakhutdinov, “Reducing the Dimensionality of Data with Neural Networks” Deep Learning history: 2012 2012: Alexnet SOTA on Imagenet Fully supervised training Deep Learning Recipe 1. Get a massive, labeled dataset 퐷 = {(푥, 푦)}: – Comp. vision: Imagenet, 1M images – Machine translation: EuroParlamanet data, CommonCrawl, several million sent. pairs – Speech recognition: 1000h (LibriSpeech), 12000h (Google Voice Search) – Question answering: SQuAD, 150k questions with human answers – … 2. Train model to maximize log 푝(푦|푥) Value of Labeled Data • Labeled data is crucial for deep learning • But labels carry little information: – Example: An ImageNet model has 30M weights, but ImageNet is about 1M images from 1000 classes Labels: 1M * 10bit = 10Mbits Raw data: (128 x 128 images): ca 500 Gbits! Value of Unlabeled Data “The brain has about 1014 synapses and we only live for about 109 seconds. So we have a lot more parameters than data. This motivates the idea that we must do a lot of unsupervised learning since the perceptual input (including proprioception) is the only place we can get 105 dimensions of constraint per second.” Geoff Hinton https://www.reddit.com/r/MachineLearning/comments/2lmo0l/ama_geoffrey_hinton/ Unsupervised learning recipe 1. Get a massive labeled dataset 퐷 = 푥 Easy, unlabeled data is nearly free 2. Train model to…??? What is the task? What is the loss function? Unsupervised learning by modeling data distribution Train the model to minimize − log 푝(푥) E.g.
    [Show full text]
  • Real-Time Black-Box Modelling with Recurrent Neural Networks
    Proceedings of the 22nd International Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, September 2–6, 2019 REAL-TIME BLACK-BOX MODELLING WITH RECURRENT NEURAL NETWORKS Alec Wright, Eero-Pekka Damskägg, and Vesa Välimäki∗ Acoustics Lab, Department of Signal Processing and Acoustics Aalto University Espoo, Finland [email protected] ABSTRACT tube amplifiers and distortion pedals. In [14] it was shown that the WaveNet model of several distortion effects was capable of This paper proposes to use a recurrent neural network for black- running in real time. The resulting deep neural network model, box modelling of nonlinear audio systems, such as tube amplifiers however, was still fairly computationally expensive to run. and distortion pedals. As a recurrent unit structure, we test both Long Short-Term Memory and a Gated Recurrent Unit. We com- In this paper, we propose an alternative black-box model based pare the proposed neural network with a WaveNet-style deep neu- on an RNN. We demonstrate that the trained RNN model is capa- ral network, which has been suggested previously for tube ampli- ble of achieving the accuracy of the WaveNet model, whilst re- fier modelling. The neural networks are trained with several min- quiring considerably less processing power to run. The proposed utes of guitar and bass recordings, which have been passed through neural network, which consists of a single recurrent layer and a the devices to be modelled. A real-time audio plugin implement- fully connected layer, is suitable for real-time emulation of tube ing the proposed networks has been developed in the JUCE frame- amplifiers and distortion pedals.
    [Show full text]
  • Approaching Hanabi with Q-Learning and Evolutionary Algorithm
    St. Cloud State University theRepository at St. Cloud State Culminating Projects in Computer Science and Department of Computer Science and Information Technology Information Technology 12-2020 Approaching Hanabi with Q-Learning and Evolutionary Algorithm Joseph Palmersten [email protected] Follow this and additional works at: https://repository.stcloudstate.edu/csit_etds Part of the Computer Sciences Commons Recommended Citation Palmersten, Joseph, "Approaching Hanabi with Q-Learning and Evolutionary Algorithm" (2020). Culminating Projects in Computer Science and Information Technology. 34. https://repository.stcloudstate.edu/csit_etds/34 This Starred Paper is brought to you for free and open access by the Department of Computer Science and Information Technology at theRepository at St. Cloud State. It has been accepted for inclusion in Culminating Projects in Computer Science and Information Technology by an authorized administrator of theRepository at St. Cloud State. For more information, please contact [email protected]. Approaching Hanabi with Q-Learning and Evolutionary Algorithm by Joseph A Palmersten A Starred Paper Submitted to the Graduate Faculty of St. Cloud State University In Partial Fulfillment of the Requirements for the Degree of Master of Science in Computer Science December, 2020 Starred Paper Committee: Bryant Julstrom, Chairperson Donald Hamnes Jie Meichsner 2 Abstract Hanabi is a cooperative card game with hidden information that requires cooperation and communication between the players. For a machine learning agent to be successful at the Hanabi, it will have to learn how to communicate and infer information from the communication of other players. To approach the problem of Hanabi the machine learning methods of Q- learning and Evolutionary algorithm are proposed as potential solutions.
    [Show full text]
  • Lecture 11 Recurrent Neural Networks I CMSC 35246: Deep Learning
    Lecture 11 Recurrent Neural Networks I CMSC 35246: Deep Learning Shubhendu Trivedi & Risi Kondor University of Chicago May 01, 2017 Lecture 11 Recurrent Neural Networks I CMSC 35246 Introduction Sequence Learning with Neural Networks Lecture 11 Recurrent Neural Networks I CMSC 35246 Some Sequence Tasks Figure credit: Andrej Karpathy Lecture 11 Recurrent Neural Networks I CMSC 35246 MLPs only accept an input of fixed dimensionality and map it to an output of fixed dimensionality Great e.g.: Inputs - Images, Output - Categories Bad e.g.: Inputs - Text in one language, Output - Text in another language MLPs treat every example independently. How is this problematic? Need to re-learn the rules of language from scratch each time Another example: Classify events after a fixed number of frames in a movie Need to resuse knowledge about the previous events to help in classifying the current. Problems with MLPs for Sequence Tasks The "API" is too limited. Lecture 11 Recurrent Neural Networks I CMSC 35246 Great e.g.: Inputs - Images, Output - Categories Bad e.g.: Inputs - Text in one language, Output - Text in another language MLPs treat every example independently. How is this problematic? Need to re-learn the rules of language from scratch each time Another example: Classify events after a fixed number of frames in a movie Need to resuse knowledge about the previous events to help in classifying the current. Problems with MLPs for Sequence Tasks The "API" is too limited. MLPs only accept an input of fixed dimensionality and map it to an output of fixed dimensionality Lecture 11 Recurrent Neural Networks I CMSC 35246 Bad e.g.: Inputs - Text in one language, Output - Text in another language MLPs treat every example independently.
    [Show full text]
  • A Guide to Recurrent Neural Networks and Backpropagation
    A guide to recurrent neural networks and backpropagation Mikael Bod´en¤ [email protected] School of Information Science, Computer and Electrical Engineering Halmstad University. November 13, 2001 Abstract This paper provides guidance to some of the concepts surrounding recurrent neural networks. Contrary to feedforward networks, recurrent networks can be sensitive, and be adapted to past inputs. Backpropagation learning is described for feedforward networks, adapted to suit our (probabilistic) modeling needs, and extended to cover recurrent net- works. The aim of this brief paper is to set the scene for applying and understanding recurrent neural networks. 1 Introduction It is well known that conventional feedforward neural networks can be used to approximate any spatially finite function given a (potentially very large) set of hidden nodes. That is, for functions which have a fixed input space there is always a way of encoding these functions as neural networks. For a two-layered network, the mapping consists of two steps, y(t) = G(F (x(t))): (1) We can use automatic learning techniques such as backpropagation to find the weights of the network (G and F ) if sufficient samples from the function is available. Recurrent neural networks are fundamentally different from feedforward architectures in the sense that they not only operate on an input space but also on an internal state space – a trace of what already has been processed by the network. This is equivalent to an Iterated Function System (IFS; see (Barnsley, 1993) for a general introduction to IFSs; (Kolen, 1994) for a neural network perspective) or a Dynamical System (DS; see e.g.
    [Show full text]
  • Feedforward Neural Networks and Word Embeddings
    Feedforward Neural Networks and Word Embeddings Fabienne Braune1 1LMU Munich May 14th, 2017 Fabienne Braune (CIS) Feedforward Neural Networks and Word Embeddings May 14th, 2017 · 1 Outline 1 Linear models 2 Limitations of linear models 3 Neural networks 4 A neural language model 5 Word embeddings Fabienne Braune (CIS) Feedforward Neural Networks and Word Embeddings May 14th, 2017 · 2 Linear Models Fabienne Braune (CIS) Feedforward Neural Networks and Word Embeddings May 14th, 2017 · 3 Binary Classification with Linear Models Example: the seminar at < time > 4 pm will Classification task: Do we have an < time > tag in the current position? Word Lemma LexCat Case SemCat Tag the the Art low seminar seminar Noun low at at Prep low stime 4 4 Digit low pm pm Other low timeid will will Verb low Fabienne Braune (CIS) Feedforward Neural Networks and Word Embeddings May 14th, 2017 · 4 Feature Vector Encode context into feature vector: 1 bias term 1 2 -3 lemma the 1 3 -3 lemma giraffe 0 ... ... ... 102 -2 lemma seminar 1 103 -2 lemma giraffe 0 ... ... ... 202 -1 lemma at 1 203 -1 lemma giraffe 0 ... ... ... 302 +1 lemma 4 1 303 +1 lemma giraffe 0 ... ... ... Fabienne Braune (CIS) Feedforward Neural Networks and Word Embeddings May 14th, 2017 · 5 Dot product with (initial) weight vector 2 3 2 3 x0 = 1 w0 = 1:00 6 x = 1 7 6 w = 0:01 7 6 1 7 6 1 7 6 x = 0 7 6 w = 0:01 7 6 2 7 6 2 7 6 ··· 7 6 ··· 7 6 7 6 7 6x = 17 6x = 0:017 6 101 7 6 101 7 6 7 6 7 6x102 = 07 6x102 = 0:017 T 6 7 6 7 h(X )= X · Θ X = 6 ··· 7 Θ = 6 ··· 7 6 7 6 7 6x201 = 17 6x201 = 0:017 6 7
    [Show full text]