Variational Autoencoder (VAE) is widely used as a generative model to approximate a model's posterior on latent variables by combining the amortized variational inference and deep neural networks. However, when paired with strong autoregressive decoders, VAE often converges to a degenerated local optimum known as "posterior collapse". Previous approaches consider the Kullback Leibler divergence (KL) individual for each datapoint. We propose to let the KL follow a distribution across the whole dataset, and analyze that it is sufficient to prevent posterior collapse by keeping the expectation of the KL's distribution positive. Then we propose Batch Normalized-VAE (BN-VAE), a simple but effective approach to set a lower bound of the expectation by regularizing the distribution of the approximate posterior's parameters. Without introducing any new model component or modifying the objective, our approach can avoid the posterior collapse effectively and efficiently. We further show that the proposed BN-VAE can be extended to conditional VAE (CVAE). Empirically, our approach surpasses strong autoregressive baselines on language modeling, text classification and dialogue generation, and rivals more complex approaches while keeping almost the same training time as VAE.
Batch Normalization (BN) is essential to effectively train state-of-the-art deep Convolutional Neural Networks (CNN). It normalizes inputs to the layers during training using the statistics of each mini-batch. In this work, we study BN from the viewpoint of Fisher kernels. We show that assuming samples within a mini-batch are from the same probability density function, then BN is identical to the Fisher vector of a Gaussian distribution. That means BN can be explained in terms of kernels that naturally emerge from the probability density function of the underlying data distribution. However, g…
Batch Normalization is quite effective at accelerating and improving the training of deep models. However, its effectiveness diminishes when the training minibatches are small, or do not consist of independent samples. We hypothesize that this is due to the dependence of model layer inputs on all the examples in the minibatch, and different activations being produced between training and inference. We propose Batch Renormalization, a simple and effective extension to ensure that the training and inference models generate the same outputs that depend on individual examples rather than the entir…
Montreal Institute for Learning Algorithms (MILA)（蒙特利尔学习算法研究所（MILA））
Improving distant speech recognition is a crucial step towards flexible human-machine interfaces. Current technology, however, still exhibits a lack of robustness, especially when adverse acoustic conditions are met. Despite the significant progress made in the last years on both speech enhancement and speech recognition, one potential limitation of state-of-the-art technology lies in composing modules that are not well matched because they are not trained jointly. To address this concern, a promising approach consists in concatenating a speech enhancement and a speech recognition deep neural …
Gradient control plays an important role in feed-forward networks applied to various computer vision tasks. Previous work has shown that Recurrent Highway Networks minimize the problem of vanishing or exploding gradients. They achieve this by setting the eigenvalues of the temporal Jacobian to 1 across the time steps. In this work, batch normalized recurrent highway networks are proposed to control the gradient flow in an improved way for network convergence. Specifically, the introduced model can be formed by batch normalizing the inputs at each recurrence loop. The proposed model is tested o…
We conduct mathematical analysis on the effect of batch normalization (BN) on gradient backpropogation in residual network training, which is believed to play a critical role in addressing the gradient vanishing/explosion problem, in this work. By analyzing the mean and variance behavior of the input and the gradient in the forward and backward passes through the BN and residual branches, respectively, we show that they work together to confine the gradient variance to a certain range across residual blocks in backpropagation. As a result, the gradient vanishing/explosion problem is avoided. W…
Université de Montréal（蒙特利尔大学）；University of Southern California（南加州大学）
Recurrent Neural Networks (RNNs) are powerful models for sequential data that have the potential to learn long-term dependencies. However, they are computationally expensive to train and difficult to parallelize. Recent work has shown that normalizing intermediate representations of neural networks can significantly improve convergence rates in feedforward neural networks . In particular, batch normalization, which uses mini-batch statistics to standardize features, was shown to significantly reduce training time. In this paper, we show that applying batch normalization to the hidden-to-hidden…
Microsoft Research AI（微软研究院人工智能）；Google Brain（谷歌大脑）
We develop a mean field theory for batch normalization in fully-connected feedforward neural networks. In so doing, we provide a precise characterization of signal propagation and gradient backpropagation in wide batch-normalized networks at initialization. Our theory shows that gradient signals grow exponentially in depth and that these exploding gradients cannot be eliminated by tuning the initial weight variances or by adjusting the nonlinear activation function. Indeed, batch normalization itself is the cause of gradient explosion. As a result, vanilla batch-normalized networks without ski…
Self Normalizing Neural Networks(SNN) proposed on Feed Forward Neural Networks(FNN) outperform regular FNN architectures in various machine learning tasks. Particularly in the domain of Computer Vision, the activation function Scaled Exponential Linear Units (SELU) proposed for SNNs, perform better than other non linear activations such as ReLU. The goal of SNN is to produce a normalized output for a normalized input. Established neural network architectures like feed forward networks and Convolutional Neural Networks(CNN) lack the intrinsic nature of normalizing outputs. Hence, requiring addi…
Entity alignment (EA) aims at finding equivalent entities in different knowledge graphs (KGs). Embedding-based approaches have dominated the EA task in recent years. Those methods face problems that come from the geometric properties of embedding vectors, including hubness and isolation. To solve these geometric problems, many normalization approaches have been adopted for EA. However, the increasing scale of KGs renders it hard for EA models to adopt the normalization processes, thus limiting their usage in real-world applications. To tackle this challenge, we present ClusterEA, a general fra…
This paper reports a novel deep architecture referred to as Maxout network In Network (MIN), which can enhance model discriminability and facilitate the process of information abstraction within the receptive field. The proposed network adopts the framework of the recently developed Network In Network structure, which slides a universal approximator, multilayer perceptron (MLP) with rectifier units, to exact features. Instead of MLP, we employ maxout MLP to learn a variety of piecewise linear activation functions and to mediate the problem of vanishing gradients that can occur when using recti…
Autodesk AI Lab（欧特克人工智能实验室）；Stanford University（斯坦福大学）
Batch normalization (BN) is a ubiquitous technique for training deep neural networks that accelerates their convergence to reach higher accuracy. However, we demonstrate that BN comes with a fundamental drawback: it incentivizes the model to rely on low-variance features that are highly specific to the training (in-domain) data, hurting generalization performance on out-of-domain examples. In this work, we investigate this phenomenon by first showing that removing BN layers across a wide range of architectures leads to lower out-of-domain and corruption errors at the cost of higher in-domain e…
Batch normalization dramatically increases the largest trainable depth of residual networks, and this benefit has been crucial to the empirical success of deep residual networks on a wide range of benchmarks. We show that this key benefit arises because, at initialization, batch normalization downscales the residual branch relative to the skip connection, by a normalizing factor on the order of the square root of the network depth. This ensures that, early in training, the function computed by normalized residual blocks in deep networks is close to the identity function (on average). We use th…
We investigate how the final parameters found by stochastic gradient descent are influenced by over-parameterization. We generate families of models by increasing the number of channels in a base network, and then perform a large hyper-parameter search to study how the test error depends on learning rate, batch size, and network width. We find that the optimal SGD hyper-parameters are determined by a "normalized noise scale," which is a function of the batch size, learning rate, and initialization conditions. In the absence of batch normalization, the optimal normalized noise scale is directly…
Indian Institute of Technology Roorkee（印度理工学院鲁基分校）
In this report, we introduce a video hashing method for scalable video segment copy detection. The objective of video segment copy detection is to find the video (s) present in a large database, one of whose segments (cropped in time) is a (transformed) copy of the given query video. This transformation may be temporal (for example frame dropping, change in frame rate) or spatial (brightness and contrast change, addition of noise etc.) in nature although the primary focus of this report is detecting temporal attacks. The video hashing method proposed by us uses a deep learning neural network t…
Off-policy reinforcement learning algorithms promise to be applicable in settings where only a fixed data-set (batch) of environment interactions is available and no new experience can be acquired. This property makes these algorithms appealing for real world problems such as robot control. In practice, however, standard off-policy algorithms fail in the batch setting for continuous control. In this paper, we propose a simple solution to this problem. It admits the use of data generated by arbitrary behavior policies and uses a learned prior -- the advantage-weighted behavior model (ABM) -- to…
Training Deep Neural Networks is complicated by the fact that the distribution of each layer's inputs changes during training, as the parameters of the previous layers change. This slows down the training by requiring lower learning rates and careful parameter initialization, and makes it notoriously hard to train models with saturating nonlinearities. We refer to this phenomenon as internal covariate shift, and address the problem by normalizing layer inputs. Our method draws its strength from making normalization a part of the model architecture and performing the normalization for each trai…
Online Normalization is a new technique for normalizing the hidden activations of a neural network. Like Batch Normalization, it normalizes the sample dimension. While Online Normalization does not use batches, it is as accurate as Batch Normalization. We resolve a theoretical limitation of Batch Normalization by introducing an unbiased technique for computing the gradient of normalized activations. Online Normalization works with automatic differentiation by adding statistical normalization as a primitive. This technique can be used in cases not covered by some other normalizers, such as recu…
Batch Normalization (BatchNorm) is commonly used in Convolutional Neural Networks (CNNs) to improve training speed and stability. However, there is still limited consensus on why this technique is effective. This paper uses concepts from the traditional adaptive filter domain to provide insight into the dynamics and inner workings of BatchNorm. First, we show that the convolution weight updates have natural modes whose stability and convergence speed are tied to the eigenvalues of the input autocorrelation matrices, which are controlled by BatchNorm through the convolution layers' channel-wise…
Much recent research has been dedicated to improving the efficiency of training and inference for image classification. This effort has commonly focused on explicitly improving theoretical efficiency, often measured as ImageNet validation accuracy per FLOP. These theoretical savings have, however, proven challenging to achieve in practice, particularly on high-performance training accelerators. In this work, we focus on improving the practical efficiency of the state-of-the-art EfficientNet models on a new class of accelerator, the Graphcore IPU. We do this by extending this family of models i…
In real-world time series recognition applications, it is possible to have data with varying length patterns. However, when using artificial neural networks (ANN), it is standard practice to use fixed-sized mini-batches. To do this, time series data with varying lengths are typically normalized so that all the patterns are the same length. Normally, this is done using zero padding or truncation without much consideration. We propose a novel method of normalizing the lengths of the time series in a dataset by exploiting the dynamic matching ability of Dynamic Time Warping (DTW). In this way, th…
Carnegie Mellon University（卡内基梅隆大学）；Beihang University（北京航空航天大学）；The University of Iowa（爱荷华大学）
In this paper, we propose a generic and simple strategy for utilizing stochastic gradient information in optimization. The technique essentially contains two consecutive steps in each iteration: 1) computing and normalizing each block (layer) of the mini-batch stochastic gradient; 2) selecting appropriate step size to update the decision variable (parameter) towards the negative of the block-normalized gradient. We conduct extensive empirical studies on various non-convex neural network optimization problems, including multi-layer perceptron, convolution neural networks and recurrent neural ne…
We revisit fuzzy neural network with a cornerstone notion of generalized hamming distance, which provides a novel and theoretically justified framework to re-interpret many useful neural network techniques in terms of fuzzy logic. In particular, we conjecture and empirically illustrate that, the celebrated batch normalization (BN) technique actually adapts the normalized bias such that it approximates the rightful bias induced by the generalized hamming distance. Once the due bias is enforced analytically, neither the optimization of bias terms nor the sophisticated batch normalization is need…
Salesforce Research（赛富时研究院）；Barcelona Supercomputing Center（巴塞罗那超级计算中心）；Montréal Institute for Learning Algorithms（蒙特利尔学习算法研究所）；Université de Montréal（蒙特利尔大学）
Residual networks (ResNet) and weight normalization play an important role in various deep learning applications. However, parameter initialization strategies have not been studied previously for weight normalized networks and, in practice, initialization methods designed for un-normalized networks are used as a proxy. Similarly, initialization for ResNets have also been studied for un-normalized networks and often under simplified settings ignoring the shortcut connection. To address these issues, we propose a novel parameter initialization strategy that avoids explosion/vanishment of informa…
Laboratory for Information and Inference Systems (LIONS), EPFL（洛桑联邦理工学院信息与推理系统实验室（LIONS））；Laboratoire Traitement et Communication d’Information, Télécom Paris, Institut Polytechnique de Paris（巴黎理工学院巴黎电信学院信息处理与通信实验室）；Department of Electrical Engineering (ESAT-STADIUS), KU Leuven（荷语鲁汶大学电气工程系（ESAT-STADIUS））
This paper introduces a family of stochastic extragradient-type algorithms for a class of nonconvex-nonconcave problems characterized by the weak Minty variational inequality (MVI). Unlike existing results on extragradient methods in the monotone setting, employing diminishing stepsizes is no longer possible in the weak MVI setting. This has led to approaches such as increasing batch sizes per iteration which can however be prohibitively expensive. In contrast, our proposed methods involves two stepsizes and only requires one additional oracle evaluation per iteration. We show that it is possi…
The Chinese University of Hong Kong（香港中文大学）；Huawei Noah’s Ark Lab（华为诺亚方舟实验室）；The University of Hong Kong（香港大学）
This work presents a probabilistic channel pruning method to accelerate Convolutional Neural Networks (CNNs). Previous pruning methods often zero out unimportant channels in training in a deterministic manner, which reduces CNN's learning capacity and results in suboptimal performance. To address this problem, we develop a probability-based pruning algorithm, called batch whitening channel pruning (BWCP), which can stochastically discard unimportant channels by modeling the probability of a channel being activated. BWCP has several merits. (1) It simultaneously trains and prunes CNNs from scra…
The most straightforward method to accelerate Stochastic Gradient Descent (SGD) computation is to distribute the randomly selected batch of inputs over multiple processors. To keep the distributed processors fully utilized requires commensurately growing the batch size. However, large batch training often leads to poorer generalization. A recently proposed solution for this problem is to use adaptive batch sizes in SGD. In this case, one starts with a small number of processes and scales the processes as training progresses. Two major challenges with this approach are (i) that dynamically resi…
University of California, Berkeley（加州大学伯克利分校）；Google AI（谷歌人工智能）
A significant advance in accelerating neural network training has been the development of normalization methods, permitting the training of deep models both faster and with better accuracy. These advances come with practical challenges: for instance, batch normalization ties the prediction of individual examples with other examples within a batch, resulting in a network that is heavily dependent on batch size. Layer normalization and group normalization are data-dependent and thus must be continually used, even at test-time. To address the issues that arise from using explicit normalization te…
While the authors of Batch Normalization (BN) identify and address an important problem involved in training deep networks-- Internal Covariate Shift-- the current solution has certain drawbacks. Specifically, BN depends on batch statistics for layerwise input normalization during training which makes the estimates of mean and standard deviation of input (distribution) to hidden layers inaccurate for validation due to shifting parameter values (especially during initial training epochs). Also, BN cannot be used with batch-size 1 during training. We address these drawbacks by proposing a non-ad…
UCLA（加州大学洛杉矶分校）；Max Planck Institute for Mathematics in the Sciences（马克斯·普朗克数学研究所）
The success of deep neural networks is in part due to the use of normalization layers. Normalization layers like Batch Normalization, Layer Normalization and Weight Normalization are ubiquitous in practice, as they improve generalization performance and speed up training significantly. Nonetheless, the vast majority of current deep learning theory and non-convex optimization literature focuses on the un-normalized setting, where the functions under consideration do not exhibit the properties of commonly normalized neural networks. In this paper, we bridge this gap by giving the first global co…
Synchronous strategies with data parallelism, such as the Synchronous StochasticGradient Descent (S-SGD) and the model averaging methods, are widely utilizedin distributed training of Deep Neural Networks (DNNs), largely owing to itseasy implementation yet promising performance. Particularly, each worker ofthe cluster hosts a copy of the DNN and an evenly divided share of the datasetwith the fixed mini-batch size, to keep the training of DNNs convergence. In thestrategies, the workers with different computational capability, need to wait foreach other because of the synchronization and delays …
In learning-to-learn the goal is to infer a learning algorithm that works well on a class of tasks sampled from an unknown meta distribution. In contrast to previous work on batch learning-to-learn, we consider a scenario where tasks are presented sequentially and the algorithm needs to adapt incrementally to improve its performance on future tasks. Key to this setting is for the algorithm to rapidly incorporate new observations into the model as they arrive, without keeping them in memory. We focus on the case where the underlying algorithm is ridge regression parameterized by a positive semi…
Imperial College London（帝国理工学院）；Shanghai Jiao Tong University（上海交通大学）
Normalization layers are essential in a Deep Convolutional Neural Network (DCNN). Various normalization methods have been proposed. The statistics used to normalize the feature maps can be computed at batch, channel, or instance level. However, in most of existing methods, the normalization for each layer is fixed. Batch-Instance Normalization (BIN) is one of the first proposed methods that combines two different normalization methods and achieve diverse normalization for different layers. However, two potential issues exist in BIN: first, the Clip function is not differentiable at input value…
