We propose the Gaussian Error Linear Unit (GELU), a high-performing neural network activation function. The GELU nonlinearity is the expected transformation of a stochastic regularizer which randomly applies the identity or zero map to a neuron's input. The GELU nonlinearity weights inputs by their magnitude, rather than gates inputs by their sign as in ReLUs. We perform an empirical evaluation of the GELU nonlinearity against the ReLU and ELU activations and find performance improvements across all considered computer vision, natural language processing, and speech tasks.
相關內容
Attention mechanism is one of the most successful techniques in deep learning based Natural Language Processing (NLP). The transformer network architecture is completely based on attention mechanisms, and it outperforms sequence-to-sequence models in neural machine translation without recurrent and convolutional layers. Grapheme-to-phoneme (G2P) conversion is a task of converting letters (grapheme sequence) to their pronunciations (phoneme sequence). It plays a significant role in text-to-speech (TTS) and automatic speech recognition (ASR) systems. In this paper, we investigate the application of transformer architecture to G2P conversion and compare its performance with recurrent and convolutional neural network based approaches. Phoneme and word error rates are evaluated on the CMUDict dataset for US English and the NetTalk dataset. The results show that transformer based G2P outperforms the convolutional-based approach in terms of word error rate and our results significantly exceeded previous recurrent approaches (without attention) regarding word and phoneme error rates on both datasets. Furthermore, the size of the proposed model is much smaller than the size of the previous approaches.
Long Short-Term Memory (LSTM) infers the long term dependency through a cell state maintained by the input and the forget gate structures, which models a gate output as a value in [0,1] through a sigmoid function. However, due to the graduality of the sigmoid function, the sigmoid gate is not flexible in representing multi-modality or skewness. Besides, the previous models lack modeling on the correlation between the gates, which would be a new method to adopt inductive bias for a relationship between previous and current input. This paper proposes a new gate structure with the bivariate Beta distribution. The proposed gate structure enables probabilistic modeling on the gates within the LSTM cell so that the modelers can customize the cell state flow with priors and distributions. Moreover, we theoretically show the higher upper bound of the gradient compared to the sigmoid function, and we empirically observed that the bivariate Beta distribution gate structure provides higher gradient values in training. We demonstrate the effectiveness of bivariate Beta gate structure on the sentence classification, image classification, polyphonic music modeling, and image caption generation.
Transformer based architectures have become de-facto models used for a range of Natural Language Processing tasks. In particular, the BERT based models achieved significant accuracy gain for GLUE tasks, CoNLL-03 and SQuAD. However, BERT based models have a prohibitive memory footprint and latency. As a result, deploying BERT based models in resource constrained environments has become a challenging task. In this work, we perform an extensive analysis of fine-tuned BERT models using second order Hessian information, and we use our results to propose a novel method for quantizing BERT models to ultra low precision. In particular, we propose a new group-wise quantization scheme, and we use a Hessian based mix-precision method to compress the model further. We extensively test our proposed method on BERT downstream tasks of SST-2, MNLI, CoNLL-03, and SQuAD. We can achieve comparable performance to baseline with at most $2.3\%$ performance degradation, even with ultra-low precision quantization down to 2 bits, corresponding up to $13\times$ compression of the model parameters, and up to $4\times$ compression of the embedding table as well as activations. Among all tasks, we observed the highest performance loss for BERT fine-tuned on SQuAD. By probing into the Hessian based analysis as well as visualization, we show that this is related to the fact that current training/fine-tuning strategy of BERT does not converge for SQuAD.
Building correspondences across different modalities, such as video and language, has recently become critical in many visual recognition applications, such as video captioning. Inspired by machine translation, recent models tackle this task using an encoder-decoder strategy. The (video) encoder is traditionally a Convolutional Neural Network (CNN), while the decoding (for language generation) is done using a Recurrent Neural Network (RNN). Current state-of-the-art methods, however, train encoder and decoder separately. CNNs are pretrained on object and/or action recognition tasks and used to encode video-level features. The decoder is then optimised on such static features to generate the video's description. This disjoint setup is arguably sub-optimal for input (video) to output (description) mapping. In this work, we propose to optimise both encoder and decoder simultaneously in an end-to-end fashion. In a two-stage training setting, we first initialise our architecture using pre-trained encoders and decoders -- then, the entire network is trained end-to-end in a fine-tuning stage to learn the most relevant features for video caption generation. In our experiments, we use GoogLeNet and Inception-ResNet-v2 as encoders and an original Soft-Attention (SA-) LSTM as a decoder. Analogously to gains observed in other computer vision problems, we show that end-to-end training significantly improves over the traditional, disjoint training process. We evaluate our End-to-End (EtENet) Networks on the Microsoft Research Video Description (MSVD) and the MSR Video to Text (MSR-VTT) benchmark datasets, showing how EtENet achieves state-of-the-art performance across the board.
For neural networks (NNs) with rectified linear unit (ReLU) or binary activation functions, we show that their training can be accomplished in a reduced parameter space. Specifically, the weights in each neuron can be trained on the unit sphere, as opposed to the entire space, and the threshold can be trained in a bounded interval, as opposed to the real line. We show that the NNs in the reduced parameter space are mathematically equivalent to the standard NNs with parameters in the whole space. The reduced parameter space shall facilitate the optimization procedure for the network training, as the search space becomes (much) smaller. We demonstrate the improved training performance using numerical examples.
Stochastic gradient Markov chain Monte Carlo (SGMCMC) has become a popular method for scalable Bayesian inference. These methods are based on sampling a discrete-time approximation to a continuous time process, such as the Langevin diffusion. When applied to distributions defined on a constrained space, such as the simplex, the time-discretisation error can dominate when we are near the boundary of the space. We demonstrate that while current SGMCMC methods for the simplex perform well in certain cases, they struggle with sparse simplex spaces; when many of the components are close to zero. However, most popular large-scale applications of Bayesian inference on simplex spaces, such as network or topic models, are sparse. We argue that this poor performance is due to the biases of SGMCMC caused by the discretization error. To get around this, we propose the stochastic CIR process, which removes all discretization error and we prove that samples from the stochastic CIR process are asymptotically unbiased. Use of the stochastic CIR process within a SGMCMC algorithm is shown to give substantially better performance for a topic model and a Dirichlet process mixture model than existing SGMCMC approaches.
We propose an architecture for VQA which utilizes recurrent layers to generate visual and textual attention. The memory characteristic of the proposed recurrent attention units offers a rich joint embedding of visual and textual features and enables the model to reason relations between several parts of the image and question. Our single model outperforms the first place winner on the VQA 1.0 dataset, performs within margin to the current state-of-the-art ensemble model. We also experiment with replacing attention mechanisms in other state-of-the-art models with our implementation and show increased accuracy. In both cases, our recurrent attention mechanism improves performance in tasks requiring sequential or relational reasoning on the VQA dataset.
We improve automatic correction of grammatical, orthographic, and collocation errors in text using a multilayer convolutional encoder-decoder neural network. The network is initialized with embeddings that make use of character N-gram information to better suit this task. When evaluated on common benchmark test data sets (CoNLL-2014 and JFLEG), our model substantially outperforms all prior neural approaches on this task as well as strong statistical machine translation-based systems with neural and task-specific features trained on the same data. Our analysis shows the superiority of convolutional neural networks over recurrent neural networks such as long short-term memory (LSTM) networks in capturing the local context via attention, and thereby improving the coverage in correcting grammatical errors. By ensembling multiple models, and incorporating an N-gram language model and edit features via rescoring, our novel method becomes the first neural approach to outperform the current state-of-the-art statistical machine translation-based approach, both in terms of grammaticality and fluency.
We develop an approach to risk minimization and stochastic optimization that provides a convex surrogate for variance, allowing near-optimal and computationally efficient trading between approximation and estimation error. Our approach builds off of techniques for distributionally robust optimization and Owen's empirical likelihood, and we provide a number of finite-sample and asymptotic results characterizing the theoretical performance of the estimator. In particular, we show that our procedure comes with certificates of optimality, achieving (in some scenarios) faster rates of convergence than empirical risk minimization by virtue of automatically balancing bias and variance. We give corroborating empirical evidence showing that in practice, the estimator indeed trades between variance and absolute performance on a training sample, improving out-of-sample (test) performance over standard empirical risk minimization for a number of classification problems.
The Residual Networks of Residual Networks (RoR) exhibits excellent performance in the image classification task, but sharply increasing the number of feature map channels makes the characteristic information transmission incoherent, which losses a certain of information related to classification prediction, limiting the classification performance. In this paper, a Pyramidal RoR network model is proposed by analysing the performance characteristics of RoR and combining with the PyramidNet. Firstly, based on RoR, the Pyramidal RoR network model with channels gradually increasing is designed. Secondly, we analysed the effect of different residual block structures on performance, and chosen the residual block structure which best favoured the classification performance. Finally, we add an important principle to further optimize Pyramidal RoR networks, drop-path is used to avoid over-fitting and save training time. In this paper, image classification experiments were performed on CIFAR-10/100 and SVHN datasets, and we achieved the current lowest classification error rates were 2.96%, 16.40% and 1.59%, respectively. Experiments show that the Pyramidal RoR network optimization method can improve the network performance for different data sets and effectively suppress the gradient disappearance problem in DCNN training.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191