Convolutional, Long Short-Term Memory, fully connected Deep Neural Networks
08 Sep 2015-pp 4580-4584
TL;DR: This paper takes advantage of the complementarity of CNNs, LSTMs and DNNs by combining them into one unified architecture, and finds that the CLDNN provides a 4-6% relative improvement in WER over an LSTM, the strongest of the three individual models.
Abstract: Both Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) have shown improvements over Deep Neural Networks (DNNs) across a wide variety of speech recognition tasks. CNNs, LSTMs and DNNs are complementary in their modeling capabilities, as CNNs are good at reducing frequency variations, LSTMs are good at temporal modeling, and DNNs are appropriate for mapping features to a more separable space. In this paper, we take advantage of the complementarity of CNNs, LSTMs and DNNs by combining them into one unified architecture. We explore the proposed architecture, which we call CLDNN, on a variety of large vocabulary tasks, varying from 200 to 2,000 hours. We find that the CLDNN provides a 4–6% relative improvement in WER over an LSTM, the strongest of the three individual models.
Citations
More filters
21 Mar 2016
TL;DR: This work formalizes the space of adversaries against deep neural networks (DNNs) and introduces a novel class of algorithms to craft adversarial samples based on a precise understanding of the mapping between inputs and outputs of DNNs.
Abstract: Deep learning takes advantage of large datasets and computationally efficient training algorithms to outperform other approaches at various machine learning tasks. However, imperfections in the training phase of deep neural networks make them vulnerable to adversarial samples: inputs crafted by adversaries with the intent of causing deep neural networks to misclassify. In this work, we formalize the space of adversaries against deep neural networks (DNNs) and introduce a novel class of algorithms to craft adversarial samples based on a precise understanding of the mapping between inputs and outputs of DNNs. In an application to computer vision, we show that our algorithms can reliably produce samples correctly classified by human subjects but misclassified in specific targets by a DNN with a 97% adversarial success rate while only modifying on average 4.02% of the input features per sample. We then evaluate the vulnerability of different sample classes to adversarial perturbations by defining a hardness measure. Finally, we describe preliminary work outlining defenses against adversarial samples by defining a predictive measure of distance between a benign input and a target classification.
3,114 citations
Cites methods from "Convolutional, Long Short-Term Memo..."
...accuracy rates higher than previous classification techniques. In short, DL is transforming computational processing of complex data in many domains such as vision [24], [37], speech recognition [15], [32], [33], language processing [13], financial fraud detection [23], and recently malware detection [14]. This increasing use of deep learning is creating incentives for adversaries to manipulate DNNs to ...
[...]
20 Mar 2016
TL;DR: Listen, Attend and Spell (LAS), a neural speech recognizer that transcribes speech utterances directly to characters without pronunciation models, HMMs or other components of traditional speech recognizers is presented.
Abstract: We present Listen, Attend and Spell (LAS), a neural speech recognizer that transcribes speech utterances directly to characters without pronunciation models, HMMs or other components of traditional speech recognizers. In LAS, the neural network architecture subsumes the acoustic, pronunciation and language models making it not only an end-to-end trained system but an end-to-end model. In contrast to DNN-HMM, CTC and most other models, LAS makes no independence assumptions about the probability distribution of the output character sequences given the acoustic sequence. Our system has two components: a listener and a speller. The listener is a pyramidal recurrent network encoder that accepts filter bank spectra as inputs. The speller is an attention-based recurrent network decoder that emits each character conditioned on all previous characters, and the entire acoustic sequence. On a Google voice search task, LAS achieves a WER of 14.1% without a dictionary or an external language model and 10.3% with language model rescoring over the top 32 beams. In comparison, the state-of-the-art CLDNN-HMM model achieves a WER of 8.0% on the same set.
2,279 citations
Cites background or methods from "Convolutional, Long Short-Term Memo..."
...The state-of-the-art model on this dataset is a CLDNN-HMM system that was described in [22]....
[...]
...Data augmentation was performed using a room simulator, adding different types of noise and reverberations; the noise sources were obtained from YouTube and environmental recordings of daily events [22]....
[...]
...Additionally, the LAS model does not use convolutional filters which have been reported to yield 5-7% WER relative improvement [22]....
[...]
...This increased the amount of audio data by 20 times with a SNR between 5dB and 30dB [22]....
[...]
18 Feb 2016
TL;DR: It is shown that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech-two vastly different languages, and is competitive with the transcription of human workers when benchmarked on standard datasets.
Abstract: We show that an end-to-end deep learning approach can be used to recognize either English or Mandarin Chinese speech-two vastly different languages. Because it replaces entire pipelines of hand-engineered components with neural networks, end-to-end learning allows us to handle a diverse variety of speech including noisy environments, accents and different languages. Key to our approach is our application of HPC techniques, enabling experiments that previously took weeks to now run in days. This allows us to iterate more quickly to identify superior architectures and algorithms. As a result, in several cases, our system is competitive with the transcription of human workers when benchmarked on standard datasets. Finally, using a technique called Batch Dispatch with GPUs in the data center, we show that our system can be inexpensively deployed in an online setting, delivering low latency when serving users at scale.
1,973 citations
Cites background or methods from "Convolutional, Long Short-Term Memo..."
...Additionally, we dynamically augment the dataset by adding unique noise every epoch with an SNR between 0dB and 30dB, just as in (Hannun et al., 2014a; Sainath et al., 2015)....
[...]
..., 2014) and work well with convolutional layers for the feature extraction (Sainath et al., 2015)....
[...]
TL;DR: A generic deep framework for activity recognition based on convolutional and LSTM recurrent units, which is suitable for multimodal wearable sensors, does not require expert knowledge in designing features, and explicitly models the temporal dynamics of feature activations is proposed.
Abstract: Human activity recognition (HAR) tasks have traditionally been solved using engineered features obtained by heuristic processes. Current research suggests that deep convolutional neural networks are suited to automate feature extraction from raw sensor inputs. However, human activities are made of complex sequences of motor movements, and capturing this temporal dynamics is fundamental for successful HAR. Based on the recent success of recurrent neural networks for time series domains, we propose a generic deep framework for activity recognition based on convolutional and LSTM recurrent units, which: (i) is suitable for multimodal wearable sensors; (ii) can perform sensor fusion naturally; (iii) does not require expert knowledge in designing features; and (iv) explicitly models the temporal dynamics of feature activations. We evaluate our framework on two datasets, one of which has been used in a public activity recognition challenge. Our results show that our framework outperforms competing deep non-recurrent networks on the challenge dataset by 4% on average; outperforming some of the previous reported results by up to 9%. Our results show that the framework can be applied to homogeneous sensor modalities, but can also fuse multimodal sensors to improve performance. We characterise key architectural hyperparameters’ influence on performance to provide insights about their optimisation.
1,896 citations
Cites background from "Convolutional, Long Short-Term Memo..."
...Nevertheless, such an approach is not strictly part of the architecture, and in the time series domain, we can see some examples of CNNs where not every convolutional layer is followed by a pooling operation [16]....
[...]
...The combination of CNNs and LSTMs in a unified framework has already offered state-of-the-art results in the speech recognition domain, where modelling temporal information is required [16]....
[...]
...Further improvements in terms of time series classification have been obtained in the speech recognition domain, by combining convolutional and recurrent layers in a unified deep framework, which contains either standard recurrent units [29] or LSTM cells [16]....
[...]
Posted Content•
TL;DR: In this paper, the authors formalize the space of adversaries against deep neural networks and introduce a novel class of algorithms to craft adversarial samples based on a precise understanding of the mapping between inputs and outputs of DNNs.
Abstract: Deep learning takes advantage of large datasets and computationally efficient training algorithms to outperform other approaches at various machine learning tasks. However, imperfections in the training phase of deep neural networks make them vulnerable to adversarial samples: inputs crafted by adversaries with the intent of causing deep neural networks to misclassify. In this work, we formalize the space of adversaries against deep neural networks (DNNs) and introduce a novel class of algorithms to craft adversarial samples based on a precise understanding of the mapping between inputs and outputs of DNNs. In an application to computer vision, we show that our algorithms can reliably produce samples correctly classified by human subjects but misclassified in specific targets by a DNN with a 97% adversarial success rate while only modifying on average 4.02% of the input features per sample. We then evaluate the vulnerability of different sample classes to adversarial perturbations by defining a hardness measure. Finally, we describe preliminary work outlining defenses against adversarial samples by defining a predictive measure of distance between a benign input and a target classification.
1,789 citations
References
More filters
TL;DR: A novel, efficient, gradient based method called long short-term memory (LSTM) is introduced, which can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units.
Abstract: Learning to store information over extended time intervals by recurrent backpropagation takes a very long time, mostly because of insufficient, decaying error backflow. We briefly review Hochreiter's (1991) analysis of this problem, then address it by introducing a novel, efficient, gradient based method called long short-term memory (LSTM). Truncating the gradient where this does not do harm, LSTM can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units. Multiplicative gate units learn to open and close access to the constant error flow. LSTM is local in space and time; its computational complexity per time step and weight is O. 1. Our experiments with artificial data involve local, distributed, real-valued, and noisy pattern representations. In comparisons with real-time recurrent learning, back propagation through time, recurrent cascade correlation, Elman nets, and neural sequence chunking, LSTM leads to many more successful runs, and learns much faster. LSTM also solves complex, artificial long-time-lag tasks that have never been solved by previous recurrent network algorithms.
72,897 citations
TL;DR: In this article, an effective way of initializing the weights that allows deep autoencoder networks to learn low-dimensional codes that work much better than principal components analysis as a tool to reduce the dimensionality of data is described.
Abstract: High-dimensional data can be converted to low-dimensional codes by training a multilayer neural network with a small central layer to reconstruct high-dimensional input vectors. Gradient descent can be used for fine-tuning the weights in such "autoencoder" networks, but this works well only if the initial weights are close to a good solution. We describe an effective way of initializing the weights that allows deep autoencoder networks to learn low-dimensional codes that work much better than principal components analysis as a tool to reduce the dimensionality of data.
16,717 citations
Proceedings Article•
31 Mar 2010TL;DR: The objective here is to understand better why standard gradient descent from random initialization is doing so poorly with deep neural networks, to better understand these recent relative successes and help design better algorithms in the future.
Abstract: Whereas before 2006 it appears that deep multilayer neural networks were not successfully trained, since then several algorithms have been shown to successfully train them, with experimental results showing the superiority of deeper vs less deep architectures. All these experimental results were obtained with new initialization or training mechanisms. Our objective here is to understand better why standard gradient descent from random initialization is doing so poorly with deep neural networks, to better understand these recent relative successes and help design better algorithms in the future. We first observe the influence of the non-linear activations functions. We find that the logistic sigmoid activation is unsuited for deep networks with random initialization because of its mean value, which can drive especially the top hidden layer into saturation. Surprisingly, we find that saturated units can move out of saturation by themselves, albeit slowly, and explaining the plateaus sometimes seen when training neural networks. We find that a new non-linearity that saturates less can often be beneficial. Finally, we study how activations and gradients vary across layers and during training, with the idea that training may be more difficult when the singular values of the Jacobian associated with each layer are far from 1. Based on these considerations, we propose a new initialization scheme that brings substantially faster convergence. 1 Deep Neural Networks Deep learning methods aim at learning feature hierarchies with features from higher levels of the hierarchy formed by the composition of lower level features. They include Appearing in Proceedings of the 13 International Conference on Artificial Intelligence and Statistics (AISTATS) 2010, Chia Laguna Resort, Sardinia, Italy. Volume 9 of JMLR: WC Weston et al., 2008). Much attention has recently been devoted to them (see (Bengio, 2009) for a review), because of their theoretical appeal, inspiration from biology and human cognition, and because of empirical success in vision (Ranzato et al., 2007; Larochelle et al., 2007; Vincent et al., 2008) and natural language processing (NLP) (Collobert & Weston, 2008; Mnih & Hinton, 2009). Theoretical results reviewed and discussed by Bengio (2009), suggest that in order to learn the kind of complicated functions that can represent high-level abstractions (e.g. in vision, language, and other AI-level tasks), one may need deep architectures. Most of the recent experimental results with deep architecture are obtained with models that can be turned into deep supervised neural networks, but with initialization or training schemes different from the classical feedforward neural networks (Rumelhart et al., 1986). Why are these new algorithms working so much better than the standard random initialization and gradient-based optimization of a supervised training criterion? Part of the answer may be found in recent analyses of the effect of unsupervised pretraining (Erhan et al., 2009), showing that it acts as a regularizer that initializes the parameters in a “better” basin of attraction of the optimization procedure, corresponding to an apparent local minimum associated with better generalization. But earlier work (Bengio et al., 2007) had shown that even a purely supervised but greedy layer-wise procedure would give better results. So here instead of focusing on what unsupervised pre-training or semi-supervised criteria bring to deep architectures, we focus on analyzing what may be going wrong with good old (but deep) multilayer neural networks. Our analysis is driven by investigative experiments to monitor activations (watching for saturation of hidden units) and gradients, across layers and across training iterations. We also evaluate the effects on these of choices of activation function (with the idea that it might affect saturation) and initialization procedure (since unsupervised pretraining is a particular form of initialization and it has a drastic impact).
9,500 citations
"Convolutional, Long Short-Term Memo..." refers methods in this paper
...The weights for all CNN and DNN layers are initialized using the Glorot-Bengio strategy described in [18]....
[...]
TL;DR: This article provides an overview of progress and represents the shared views of four research groups that have had recent successes in using DNNs for acoustic modeling in speech recognition.
Abstract: Most current speech recognition systems use hidden Markov models (HMMs) to deal with the temporal variability of speech and Gaussian mixture models (GMMs) to determine how well each state of each HMM fits a frame or a short window of frames of coefficients that represents the acoustic input. An alternative way to evaluate the fit is to use a feed-forward neural network that takes several frames of coefficients as input and produces posterior probabilities over HMM states as output. Deep neural networks (DNNs) that have many hidden layers and are trained using new methods have been shown to outperform GMMs on a variety of speech recognition benchmarks, sometimes by a large margin. This article provides an overview of this progress and represents the shared views of four research groups that have had recent successes in using DNNs for acoustic modeling in speech recognition.
9,091 citations
Posted Content•
TL;DR: This paper proposes a gradient norm clipping strategy to deal with exploding gradients and a soft constraint for the vanishing gradients problem and validates empirically the hypothesis and proposed solutions.
Abstract: There are two widely known issues with properly training Recurrent Neural Networks, the vanishing and the exploding gradient problems detailed in Bengio et al. (1994). In this paper we attempt to improve the understanding of the underlying issues by exploring these problems from an analytical, a geometric and a dynamical systems perspective. Our analysis is used to justify a simple yet effective solution. We propose a gradient norm clipping strategy to deal with exploding gradients and a soft constraint for the vanishing gradients problem. We validate empirically our hypothesis and proposed solutions in the experimental section.
3,549 citations
Additional excerpts
...One may argue that if the LSTMs are better initialized, such that better temporal modeling can be performed, are CNNs really necessary? Our initial experiments with LSTMs use Gaussian random weight initialization, which produces eigenvalues of the initial recurrent network which are close to zero, thus increasing the chances for vanishing gradients [19]....
[...]