We provide a theoretical explanation for the effectiveness of gradient clipping in training deep neural networks. The key ingredient is a new smoothness condition derived from practical neural network training examples. We observe that gradient smoothness, a concept central to the analysis of first-order optimization algorithms that is often assumed to be a constant, demonstrates significant variability along the training trajectory of deep neural networks. Further, this smoothness positively correlates with the gradient norm, and contrary to standard assumptions in the literature, it can grow with the norm of the gradient. These empirical observations limit the applicability of existing theoretical analyses of algorithms that rely on a fixed bound on smoothness. These observations motivate us to introduce a novel relaxation of gradient smoothness that is weaker than the commonly used Lipschitz smoothness assumption. Under the new condition, we prove that two popular methods, namely, \emph{gradient clipping} and \emph{normalized gradient}, converge arbitrarily faster than gradient descent with fixed stepsize. We further explain why such adaptively scaled gradient methods can accelerate empirical convergence and verify our results empirically in popular neural network training settings.
We investigate adversarial robustness of Gaussian Process Classification (GPC) models. Given a compact subset of the input space $T\subseteq \mathbb{R}^d$ enclosing a test point $x^*$ and a GPC trained on a dataset $\mathcal{D}$, we aim to compute the minimum and the maximum classification probability for the GPC over all the points in $T$. In order to do so, we show how functions lower- and upper-bounding the GPC output in $T$ can be derived, and implement those in a branch and bound optimisation algorithm. For any error threshold $\epsilon > 0$ selected a priori, we show that our algorithm is guaranteed to reach values $\epsilon$-close to the actual values in finitely many iterations. We apply our method to investigate the robustness of GPC models on a 2D synthetic dataset, the SPAM dataset and a subset of the MNIST dataset, providing comparisons of different GPC training techniques, and show how our method can be used for interpretability analysis. Our empirical analysis suggests that GPC robustness increases with more accurate posterior estimation.
Principal Component Analysis (PCA) has been used to study the pathogenesis of diseases. To enhance the interpretability of classical PCA, various improved PCA methods have been proposed to date. Among these, a typical method is the so-called sparse PCA, which focuses on seeking sparse loadings. However, the performance of these methods is still far from satisfactory due to their limitation of using unsupervised learning methods; moreover, the class ambiguity within the sample is high. To overcome this problem, this study developed a new PCA method, which is named the Supervised Discriminative Sparse PCA (SDSPCA). The main innovation of this method is the incorporation of discriminative information and sparsity into the PCA model. Specifically, in contrast to the traditional sparse PCA, which imposes sparsity on the loadings, here, sparse components are obtained to represent the data. Furthermore, via linear transformation, the sparse components approximate the given label information. On the one hand, sparse components improve interpretability over traditional PCA, while on the other hand, they are have discriminative abilities suitable for classification purposes. A simple algorithm is developed and its convergence proof is provided. The SDSPCA has been applied to common characteristic gene selection (com-characteristic gene) and tumor classification on multi-view biological data. The sparsity and classification performance of the SDSPCA are empirically verified via abundant, reasonable, and effective experiments, and the obtained results demonstrate that SDSPCA outperforms other state-of-the-art methods.
Deep convolutional neural networks (DCNNs) are powerful models that yield impressive results at object classification. However, recent work has shown that they do not generalize well to partially occluded objects and to mask attacks. In contrast to DCNNs, compositional models are robust to partial occlusion, however, they are not as discriminative as deep models. In this work, we combine DCNNs and compositional object models to retain the best of both approaches: a discriminative model that is robust to partial occlusion and mask attacks. Our model is learned in two steps. First, a standard DCNN is trained for image classification. Subsequently, we cluster the DCNN features into dictionaries. We show that the dictionary components resemble object part detectors and learn the spatial distribution of parts for each object class. We propose mixtures of compositional models to account for large changes in the spatial activation patterns (e.g. due to changes in the 3D pose of an object). At runtime, an image is first classified by the DCNN in a feedforward manner. The prediction uncertainty is used to detect partially occluded objects, which in turn are classified by the compositional model. Our experimental results demonstrate that combining compositional models and DCNNs resolves a fundamental problem of current deep learning approaches to computer vision: The combined model recognizes occluded objects, even when it has not been exposed to occluded objects during training, while at the same time maintaining high discriminative performance for non-occluded objects.
Appearance and motion are two key components to depict and characterize the video content. Currently, the two-stream models have achieved state-of-the-art performances on video classification. However, extracting motion information, specifically in the form of optical flow features, is extremely computationally expensive, especially for large-scale video classification. In this paper, we propose a motion hallucination network, namely MoNet, to imagine the optical flow features from the appearance features, with no reliance on the optical flow computation. Specifically, MoNet models the temporal relationships of the appearance features and exploits the contextual relationships of the optical flow features with concurrent connections. Extensive experimental results demonstrate that the proposed MoNet can effectively and efficiently hallucinate the optical flow features, which together with the appearance features consistently improve the video classification performances. Moreover, MoNet can help cutting down almost a half of computational and data-storage burdens for the two-stream video classification. Our code is available at: https://github.com/YongyiTang92/MoNet-Features.
Extreme classification seeks to assign each data point, the most relevant labels from a universe of a million or more labels. This task is faced with the dual challenge of high precision and scalability, with millisecond level prediction times being a benchmark. We propose DEFRAG, an adaptive feature agglomeration technique to accelerate extreme classification algorithms. Despite past works on feature clustering and selection, DEFRAG distinguishes itself in being able to scale to millions of features, and is especially beneficial when feature sets are sparse, which is typical of recommendation and multi-label datasets. The method comes with provable performance guarantees and performs efficient task-driven agglomeration to reduce feature dimensionalities by an order of magnitude or more. Experiments show that DEFRAG can not only reduce training and prediction times of several leading extreme classification algorithms by as much as 40%, but also be used for feature reconstruction to address the problem of missing features, as well as offer superior coverage on rare labels.
Leather is a type of natural, durable, flexible, soft, supple and pliable material with smooth texture. It is commonly used as a raw material to manufacture luxury consumer goods for high-end customers. To ensure good quality control on the leather products, one of the critical processes is the visual inspection step to spot the random defects on the leather surfaces and it is usually conducted by experienced experts. This paper presents an automatic mechanism to perform the leather defect classification. In particular, we focus on detecting tick-bite defects on a specific type of calf leather. Both the handcrafted feature extractors (i.e., edge detectors and statistical approach) and data-driven (i.e., artificial neural network) methods are utilized to represent the leather patches. Then, multiple classifiers (i.e., decision trees, Support Vector Machines, nearest neighbour and ensemble classifiers) are exploited to determine whether the test sample patches contain defective segments. Using the proposed method, we managed to get a classification accuracy rate of 84% from a sample of approximately 2500 pieces of 400 * 400 leather patches.
We present a fast and scalable algorithm to induce non-monotonic logic programs from statistical learning models. We reduce the problem of search for best clauses to instances of the High-Utility Itemset Mining (HUIM) problem. In the HUIM problem, feature values and their importance are treated as transactions and utilities respectively. We make use of TreeExplainer, a fast and scalable implementation of the Explainable AI tool SHAP, to extract locally important features and their weights from ensemble tree models. Our experiments with UCI standard benchmarks suggest a significant improvement in terms of classification evaluation metrics and running time of the training algorithm compared to ALEPH, a state-of-the-art Inductive Logic Programming (ILP) system.
Unmanned aerial vehicles (UAV) are used in precision agriculture (PA) to enable aerial monitoring of farmlands. Intelligent methods are required to pinpoint weed infestations and make optimal choice of pesticide. UAV can fly a multispectral camera and collect data. However, the classification of multispectral images using supervised machine learning algorithms such as convolutional neural networks (CNN) requires large amount of training data. This is a common drawback in deep learning we try to circumvent making use of a semi-supervised generative adversarial networks (GAN), providing a pixel-wise classification for all the acquired multispectral images. Our algorithm consists of a generator network that provides photo-realistic images as extra training data to a multi-class classifier, acting as a discriminator and trained on small amounts of labeled data. The performance of the proposed method is evaluated on the weedNet dataset consisting of multispectral crop and weed images collected by a micro aerial vehicle (MAV). The results by the proposed semi-supervised GAN achieves high classification accuracy and demonstrates the potential of GAN-based methods for the challenging task of multispectral image classification.
Tactile enhanced multimedia is generated by synchronizing traditional multimedia clips, to generate hot and cold air effect, with an electric heater and a fan. This objective is to give viewers a more realistic and immersing feel of the multimedia content. The response to this enhanced multimedia content (mulsemedia) is evaluated in terms of the appreciation/emotion by using human brain signals. We observe and record electroencephalography (EEG) data using a commercially available four channel MUSE headband. A total of 21 participants voluntarily participated in this study for EEG recordings. We extract frequency domain features from five different bands of each EEG channel. Four emotions namely: happy, relaxed, sad, and angry are classified using a support vector machine in response to the tactile enhanced multimedia. An increased accuracy of 76:19% is achieved when compared to 63:41% by using the time domain features. Our results show that the selected frequency domain features could be better suited for emotion classification in mulsemedia studies.
Domain generalization is the problem of assigning labels to an unlabeled data set, given several similar data sets for which labels have been provided. Despite considerable interest in this problem over the last decade, there has been no theoretical analysis in the setting of multi-class classification. In this work, we study a kernel-based learning algorithm and establish a generalization error bound that scales logarithmically in the number of classes, matching state-of-the-art bounds for multi-class classification in the conventional learning setting. We also demonstrate empirically that the proposed algorithm achieves significant performance gains compared to a pooling strategy.
Given a sufficiently large training dataset, it is relatively easy to train a modern convolution neural network (CNN) as a required image classifier. However, for the task of fish classification and/or fish detection, if a CNN was trained to detect or classify particular fish species in particular background habitats, the same CNN exhibits much lower accuracy when applied to new/unseen fish species and/or fish habitats. Therefore, in practice, the CNN needs to be continuously fine-tuned to improve its classification accuracy to handle new project-specific fish species or habitats. In this work we present a labelling-efficient method of training a CNN-based fish-detector (the Xception CNN was used as the base) on relatively small numbers (4,000) of project-domain underwater fish/no-fish images from 20 different habitats. Additionally, 17,000 of known negative (that is, missing fish) general-domain (VOC2012) above-water images were used. Two publicly available fish-domain datasets supplied additional 27,000 of above-water and underwater positive/fish images. By using this multi-domain collection of images, the trained Xception-based binary (fish/not-fish) classifier achieved 0.17% false-positives and 0.61% false-negatives on the project's 20,000 negative and 16,000 positive holdout test images, respectively. The area under the ROC curve (AUC) was 99.94%.
Networks have been widely used as the data structure for abstracting real-world systems as well as organizing the relations among entities. Network embedding models are powerful tools in mapping nodes in a network into continuous vector-space representations in order to facilitate subsequent tasks such as classification and link prediction. Existing network embedding models comprehensively integrate all information of each node, such as links and attributes, towards a single embedding vector to represent the node's general role in the network. However, a real-world entity could be multifaceted, where it connects to different neighborhoods due to different motives or self-characteristics that are not necessarily correlated. For example, in a movie recommender system, a user may love comedies or horror movies simultaneously, but it is not likely that these two types of movies are mutually close in the embedding space, nor the user embedding vector could be sufficiently close to them at the same time. In this paper, we propose a polysemous embedding approach for modeling multiple facets of nodes, as motivated by the phenomenon of word polysemy in language modeling. Each facet of a node is mapped as an embedding vector, while we also maintain association degree between each pair of node and facet. The proposed method is adaptive to various existing embedding models, without significantly complicating the optimization process. We also discuss how to engage embedding vectors of different facets for inference tasks including classification and link prediction. Experiments on real-world datasets help comprehensively evaluate the performance of the proposed method.
We propose a new family of fairness definitions for classification problems that combine some of the best properties of both statistical and individual notions of fairness. We posit not only a distribution over individuals, but also a distribution over (or collection of) classification tasks. We then ask that standard statistics (such as error or false positive/negative rates) be (approximately) equalized across individuals, where the rate is defined as an expectation over the classification tasks. Because we are no longer averaging over coarse groups (such as race or gender), this is a semantically meaningful individual-level constraint. Given a sample of individuals and classification problems, we design an oracle-efficient algorithm (i.e. one that is given access to any standard, fairness-free learning heuristic) for the fair empirical risk minimization task. We also show that given sufficiently many samples, the ERM solution generalizes in two directions: both to new individuals, and to new classification tasks, drawn from their corresponding distributions. Finally we implement our algorithm and empirically verify its effectiveness.
Person re-identification (Re-ID) across multiple datasets is a challenging task due to two main reasons: the presence of large cross-dataset distinctions and the absence of annotated target instances. To address these two issues, this paper proposes a domain adaptive attention learning approach to reliably transfer discriminative representation from the labeled source domain to the unlabeled target domain. In this approach, a domain adaptive attention model is learned to separate the feature map into domain-shared part and domain-specific part. In this manner, the domain-shared part is used to capture transferable cues that can compensate cross-dataset distinctions and give positive contributions to the target task, while the domain-specific part aims to model the noisy information to avoid the negative transfer caused by domain diversity. A soft label loss is further employed to take full use of unlabeled target data by estimating pseudo labels. Extensive experiments on the Market-1501, DukeMTMC-reID and MSMT17 benchmarks demonstrate the proposed approach outperforms the state-of-the-arts.
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.
Language and vision are processed as two different modal in current work for image captioning. However, recent work on Super Characters method shows the effectiveness of two-dimensional word embedding, which converts text classification problem into image classification problem. In this paper, we propose the SuperCaptioning method, which borrows the idea of two-dimensional word embedding from Super Characters method, and processes the information of language and vision together in one single CNN model. The experimental results on Flickr30k data shows the proposed method gives high quality image captions. An interactive demo is ready to show at the workshop.
Convolutional Neural Networks (CNN) have recently seen tremendous success in various computer vision tasks. However, their application to problems with high dimensional input and output, such as high-resolution image and video segmentation or 3D medical imaging, has been limited by various factors. Primarily, in the training stage, it is necessary to store network activations for back propagation. In these settings, the memory requirements associated with storing activations can exceed what is feasible with current hardware, especially for problems in 3D. Motivated by the propagation of signals over physical networks, that are governed by the hyperbolic Telegraph equation, in this work we introduce a fully conservative hyperbolic network for problems with high dimensional input and output. We introduce a coarsening operation that allows completely reversible CNNs by using a learnable Discrete Wavelet Transform and its inverse to both coarsen and interpolate the network state and change the number of channels. We show that fully reversible networks are able to achieve results comparable to the state of the art in 4D time-lapse hyper spectral image segmentation and full 3D video segmentation, with a much lower memory footprint that is a constant independent of the network depth. We also extend the use of such networks to Variational Auto Encoders with high resolution input and output.
We consider multi-label classification where the goal is to annotate each data point with the most relevant $\textit{subset}$ of labels from an extremely large label set. Efficient annotation can be achieved with balanced tree predictors, i.e. trees with logarithmic-depth in the label complexity, whose leaves correspond to labels. Designing prediction mechanism with such trees for real data applications is non-trivial as it needs to accommodate sending examples to multiple leaves while at the same time sustain high prediction accuracy. In this paper we develop the LdSM algorithm for the construction and training of multi-label decision trees, where in every node of the tree we optimize a novel objective function that favors balanced splits, maintains high class purity of children nodes, and allows sending examples to multiple directions but with a penalty that prevents tree over-growth. Each node of the tree is trained once the previous node is completed leading to a streaming approach for training. We analyze the proposed objective theoretically and show that minimizing it leads to pure and balanced data splits. Furthermore, we show a boosting theorem that captures its connection to the multi-label classification error. Experimental results on benchmark data sets demonstrate that our approach achieves high prediction accuracy and low prediction time and position LdSM as a competitive tool among existing state-of-the-art approaches.
We introduce a novel method to combat label noise when training deep neural networks for classification. We propose a loss function that permits abstention during training thereby allowing the DNN to abstain on confusing samples while continuing to learn and improve classification performance on the non-abstained samples. We show how such a deep abstaining classifier (DAC) can be used for robust learning in the presence of different types of label noise. In the case of structured or systematic label noise -- where noisy training labels or confusing examples are correlated with underlying features of the data-- training with abstention enables representation learning for features that are associated with unreliable labels. In the case of unstructured (arbitrary) label noise, abstention during training enables the DAC to be used as an effective data cleaner by identifying samples that are likely to have label noise. We provide analytical results on the loss function behavior that enable dynamic adaption of abstention rates based on learning progress during training. We demonstrate the utility of the deep abstaining classifier for various image classification tasks under different types of label noise; in the case of arbitrary label noise, we show significant improvements over previously published results on multiple image benchmarks. Source code is available at https://github.com/thulas/dac-label-noise