Document-level context has received lots of attention for compensating neural machine translation (NMT) of isolated sentences. However, recent advances in document-level NMT focus on sophisticated integration of the c...
详细信息
We propose simple architectural modifications in the standard Transformer with the goal to reduce its total state size (defined as the number of self-attention layers times the sum of the key and value dimensions, tim...
ISBN:
(数字)9781509066315
ISBN:
(纸本)9781509066322
We propose simple architectural modifications in the standard Transformer with the goal to reduce its total state size (defined as the number of self-attention layers times the sum of the key and value dimensions, times position) without loss of performance. Large scale Transformer language models have been empirically proved to give very good performance. However, scaling up results in a model that needs to store large states at evaluation time. This can increase the memory requirement dramatically for search e.g., in speech recognition (first pass decoding, lattice rescoring, or shallow fusion). In order to efficiently increase the model capacity without increasing the state size, we replace the single-layer feed-forward module in the Transformer layer by a deeper network, and decrease the total number of layers. In addition, we also evaluate the effect of key-value tying which directly divides the state size in half. On TED-LIUM 2, we obtain a model of state size 4 times smaller than the standard Transformer, with only 2% relative loss in terms of perplexity, which makes the deployment of Transformer language models more convenient.
Training deep neural networks is often challenging in terms of training stability. It often requires careful hyperparameter tuning or a pretraining scheme to converge. Layer normalization (LN) has shown to be a crucia...
ISBN:
(数字)9781509066315
ISBN:
(纸本)9781509066322
Training deep neural networks is often challenging in terms of training stability. It often requires careful hyperparameter tuning or a pretraining scheme to converge. Layer normalization (LN) has shown to be a crucial ingredient in training deep encoder-decoder models. We explore various LN long short-term memory (LSTM) recurrent neural networks (RNN) variants by applying LN to different parts of the internal recurrency of LSTMs. There is no previous work that investigates this. We carry out experiments on the Switchboard 300h task for both hybrid and end-to-end ASR models and we show that LN improves the final word error rate (WER), the stability during training, allows to train even deeper models, requires less hyperparameter tuning, and works well even without pre-training. We find that applying LN to both forward and recurrent inputs globally, which we denoted by Global Joined Norm variant, gives a 10% relative improvement in WER.
We propose several tracking adaptation approaches to recover from early tracking errors in sign languagerecognition by optimizing the obtained tracking paths w.r.t. to the hypothesized word sequences of an automatic ...
详细信息
We propose several tracking adaptation approaches to recover from early tracking errors in sign languagerecognition by optimizing the obtained tracking paths w.r.t. to the hypothesized word sequences of an automatic sign languagerecognition system. Hand or head tracking is usually only optimized according to a tracking criterion. As a consequence, methods which depend on accurate detection and tracking of body parts lead to recognition errors in gesture and sign language processing. We analyze an integrated tracking and recognition approach addressing these problems and propose approximation approaches over multiple hand hypotheses to ease the time complexity of the integrated approach. Most state-of-the-art systems consider tracking as a preprocessing feature extraction part. Experiments on a publicly available benchmark database show that the proposed methods strongly improve the recognition accuracy of the system.
Long short-term memory (LSTM) networks are the dominant architecture for large vocabulary continuous speech recognition (LVCSR) acoustic modeling due to their good performance. However, LSTMs are hard to tune and comp...
详细信息
ISBN:
(数字)9781509066315
ISBN:
(纸本)9781509066322
Long short-term memory (LSTM) networks are the dominant architecture for large vocabulary continuous speech recognition (LVCSR) acoustic modeling due to their good performance. However, LSTMs are hard to tune and computationally expensive. To build a system with lower computational costs and which allows online streaming applications, we explore convolutional neural networks (CNN). To the best of our knowledge there is no overview on CNN hyper-parameter tuning for LVCSR in the literature, so we present our results explicitly. Apart from recognition performance, we focus on the training and evaluation speed and provide a time-efficient setup for CNNs. We faced an overfitting problem in training and solved it with data augmentation, namely SpecAugment. The system achieves results competitive with the top LSTM results. We significantly increased the speed of CNN in training and decoding approaching the speed of the offline LSTM.
One of the most difficult challenges in face recognition is the large variation in pose. One approach to handle this problem is to use a 2D-Warping algorithm in a nearest-neighbor classifier. The 2D-Warping algorithm ...
详细信息
One of the most difficult challenges in face recognition is the large variation in pose. One approach to handle this problem is to use a 2D-Warping algorithm in a nearest-neighbor classifier. The 2D-Warping algorithm optimizes an energy function that captures the cost of matching pixels between two images while respecting the 2D dependencies defined by local pixel neighborhoods. Optimizing this energy function is an NP-complete problem and is therefore approached with algorithms that aim to approximate the optimal solution. In this paper we compare two algorithms that do this without discarding any 2D dependencies and we study the effect of the quality of the approximate solutions on the classification performance. Additionally, we propose a new algorithm that is capable of finding better solutions and obtaining better energies than the other methods. The experimental evaluation on the CMU-MultiPIE database shows that the proposed algorithm also achieves state-of-the-art recognition accuracies.
The task of fine-grained visual classification (FGVC) deals with classification problems that display a small inter-class variance such as distinguishing between different bird species or car models. State-of-the-art ...
详细信息
ISBN:
(数字)9781728165530
ISBN:
(纸本)9781728165547
The task of fine-grained visual classification (FGVC) deals with classification problems that display a small inter-class variance such as distinguishing between different bird species or car models. State-of-the-art approaches typically tackle this problem by integrating an elaborate attention mechanism or (part-) localization method into a standard convolutional neural network (CNN). Also in this work the aim is to enhance the performance of a backbone CNN such as ResNet by including three efficient and lightweight components specifically designed for FGVC. This is achieved by using global k-max pooling, a discriminative embedding layer trained by optimizing class means and an efficient localization module that estimates bounding boxes using only class labels for training. The resulting model achieves state-of-the-art recognition accuracies on multiple FGVC benchmark datasets.
Log-linear models are a promising approach for speech recognition. Typically, log-linear models are trained according to a strictly convex criterion. Optimization algorithms are guaranteed to converge to the unique gl...
详细信息
Log-linear models are a promising approach for speech recognition. Typically, log-linear models are trained according to a strictly convex criterion. Optimization algorithms are guaranteed to converge to the unique global optimum of the objective function from any initialization. For large-scale applications, considerations in the limit of infinite iterations are not sufficient. We show that log-linear training can be a highly ill-conditioned optimization problem, resulting in extremely slow convergence. Conversely, the optimization problem can be preconditioned by feature transformations. Making use of our convergence analysis, we improve our log-linear speech recognition system and achieve a strong reduction of its training time. In addition, we validate our analysis on a continuous handwriting recognition task.
In statistical machine translation, word lattices are used to represent the ambiguities in the preprocessing of the source sentence, such as word segmentation for Chinese or morphological analysis for German. Several ...
详细信息
ISBN:
(纸本)9781622765928
In statistical machine translation, word lattices are used to represent the ambiguities in the preprocessing of the source sentence, such as word segmentation for Chinese or morphological analysis for German. Several approaches have been proposed to define the probability of different paths through the lattice with external tools like word segmenters, or by applying indicator features. We introduce a novel lattice design, which explicitly distinguishes between different preprocessing alternatives for the source sentence. It allows us to make use of specific features for each preprocessing type and to lexicalize the choice of lattice path directly in the phrase translation model. We argue that forced alignment training can be used to learn lattice path and phrase translation model simultaneously. On the news-commentary portion of the German→English WMT 2011 task we can show moderate improvements of up to 0.6% Bleu over a state-of-the-art baseline system.
In this paper, we study a simple yet elegant latent variable attention model for automatic speech recognition (ASR) which enables an integration of attention sequence modeling into the direct hidden Markov model (HMM)...
ISBN:
(数字)9781509066315
ISBN:
(纸本)9781509066322
In this paper, we study a simple yet elegant latent variable attention model for automatic speech recognition (ASR) which enables an integration of attention sequence modeling into the direct hidden Markov model (HMM) concept. We use a sequence of hidden variables that establishes a mapping from output labels to input frames. Inspired by the direct HMM model, we assume a decomposition of the label sequence posterior into emission and transition probabilities using zero-order assumption and incorporate both Transformer and LSTM attention models into it. The method keeps the explicit alignment as part of the stochastic model and combines the ease of the end-to-end training of the attention model as well as an efficient and simple beam search. To study the effect of the latent model, we qualitatively analyze the alignment behavior of the different approaches. Our experiments on three ASR tasks show promising results in WER with more focused alignments in comparison to the attention models.
暂无评论