In recent years, deep learning has achieved remarkable achievements in many fields, including computer vision, natural language processing, speech recognition and others. Adequate training data is the key to ensure the effectiveness of the deep models. However, obtaining valid data requires a lot of time and labor resources. Data augmentation (DA) is an effective alternative approach, which can generate new labeled data based on existing data using label-preserving transformations. Although we can benefit a lot from DA, designing appropriate DA policies requires a lot of expert experience and time consumption, and the evaluation of searching the optimal policies is costly. So we raise a new question in this paper: how to achieve automated data augmentation at as low cost as possible? We propose a method named BO-Aug for automating the process by finding the optimal DA policies using the Bayesian optimization approach. Our method can find the optimal policies at a relatively low search cost, and the searched policies based on a specific dataset are transferable across different neural network architectures or even different datasets. We validate the BO-Aug on three widely used image classification datasets, including CIFAR-10, CIFAR-100 and SVHN. Experimental results show that the proposed method can achieve state-of-the-art or near advanced classification accuracy. Code to reproduce our experiments is available at https://github.com/zhangxiaozao/BO-Aug.
Semi-supervised learning is sought for leveraging the unlabelled data when labelled data is difficult or expensive to acquire. Deep generative models (e.g., Variational Autoencoder (VAE)) and semisupervised Generative Adversarial Networks (GANs) have recently shown promising performance in semi-supervised classification for the excellent discriminative representing ability. However, the latent code learned by the traditional VAE is not exclusive (repeatable) for a specific input sample, which prevents it from excellent classification performance. In particular, the learned latent representation depends on a non-exclusive component which is stochastically sampled from the prior distribution. Moreover, the semi-supervised GAN models generate data from pre-defined distribution (e.g., Gaussian noises) which is independent of the input data distribution and may obstruct the convergence and is difficult to control the distribution of the generated data. To address the aforementioned issues, we propose a novel Adversarial Variational Embedding (AVAE) framework for robust and effective semi-supervised learning to leverage both the advantage of GAN as a high quality generative model and VAE as a posterior distribution learner. The proposed approach first produces an exclusive latent code by the model which we call VAE++, and meanwhile, provides a meaningful prior distribution for the generator of GAN. The proposed approach is evaluated over four different real-world applications and we show that our method outperforms the state-of-the-art models, which confirms that the combination of VAE++ and GAN can provide significant improvements in semisupervised classification.
We consider the extreme multi-label text classification (XMC) problem: given an input text, return the most relevant labels from a large label collection. For example, the input text could be a product description on Amazon.com and the labels could be product categories. XMC is an important yet challenging problem in the NLP community. Recently, deep pretrained transformer models have achieved state-of-the-art performance on many NLP tasks including sentence classification, albeit with small label sets. However, naively applying deep transformer models to the XMC problem leads to sub-optimal performance due to the large output space and the label sparsity issue. In this paper, we propose X-Transformer, the first scalable approach to fine-tuning deep transformer models for the XMC problem. The proposed method achieves new state-of-the-art results on four XMC benchmark datasets. In particular, on a Wiki dataset with around 0.5 million labels, the prec@1 of X-Transformer is 77.28%, a substantial improvement over state-of-the-art XMC approaches Parabel (linear) and AttentionXML (neural), which achieve 68.70% and 76.95% precision@1, respectively. We further apply X-Transformer to a product2query dataset from Amazon and gained 10.7% relative improvement on prec@1 over Parabel.
Graph Neural Networks (GNNs) have proven to be successful in many classification tasks, outperforming previous state-of-the-art methods in terms of accuracy. However, accuracy alone is not enough for high-stakes decision making. Decision makers want to know the likelihood that a specific GNN prediction is correct. For this purpose, obtaining calibrated models is essential. In this work, we perform an empirical evaluation of the calibration of state-of-the-art GNNs on multiple datasets. Our experiments show that GNNs can be calibrated in some datasets but also badly miscalibrated in others, and that state-of-the-art calibration methods are helpful but do not fix the problem.
The increasing availability of sensor data at machine tools makes automatic chatter detection algorithms a trending topic in metal cutting. Two prominent and advanced methods for feature extraction via signal decomposition are Wavelet Packet Transform (WPT) and Ensemble Empirical Mode Decomposition (EEMD). We apply these two methods to time series acquired from an acceleration sensor at the tool holder of a lathe. Different turning experiments with varying dynamic behavior of the machine tool structure were performed. We compare the performance of these two methods with Support Vector Machine (SVM), Logistic Regression, Random Forest Classification and Gradient Boosting combined with Recursive Feature Elimination (RFE). We also show that the common WPT-based approach of choosing wavelet packets with the highest energy ratios as representative features for chatter does not always result in packets that enclose the chatter frequency, thus reducing the classification accuracy. Further, we test the transfer learning capability of each of these methods by training the classifier on one of the cutting configurations and then testing it on the other cases. It is found that when training and testing on data from the same cutting configuration both methods yield high accuracies reaching in one of the cases as high as 94% and 95%, respectively, for WPT and EEMD. However, our experimental results show that EEMD can outperform WPT in transfer learning applications with accuracy of up to 95%.
Multiple Instance Learning (MIL) is a weak supervision learning paradigm that allows modeling of machine learning problems in which labels are available only for groups of examples called bags. A positive bag may contain one or more positive examples but it is not known which examples in the bag are positive. All examples in a negative bag belong to the negative class. Such problems arise frequently in fields of computer vision, medical image processing and bioinformatics. Many neural network based solutions have been proposed in the literature for MIL, however, almost all of them rely on introducing specialized blocks and connectivity in the architectures. In this paper, we present a novel and effective approach to Multiple Instance Learning in neural networks. Instead of making changes to the architectures, we propose a simple bag-level ranking loss function that allows Multiple Instance Classification in any neural architecture. We have demonstrated the effectiveness of our proposed method for popular MIL benchmark datasets. In addition, we have tested the performance of our method in convolutional neural networks used to model an MIL problem derived from the well-known MNIST dataset. Results have shown that despite being simpler, our proposed scheme is comparable or better than existing methods in the literature in practical scenarios. Python code files for all the experiments can be found at https://github.com/amina01/ESMIL.
Graphs are a natural abstraction for many problems where nodes represent entities and edges represent a relationship across entities. An important area of research that has emerged over the last decade is the use of graphs as a vehicle for non-linear dimensionality reduction in a manner akin to previous efforts based on manifold learning with uses for downstream database processing, machine learning and visualization. In this systematic yet comprehensive experimental survey, we benchmark several popular network representation learning methods operating on two key tasks: link prediction and node classification. We examine the performance of 12 unsupervised embedding methods on 15 datasets. To the best of our knowledge, the scale of our study -- both in terms of the number of methods and number of datasets -- is the largest to date. Our results reveal several key insights about work-to-date in this space. First, we find that certain baseline methods (task-specific heuristics, as well as classic manifold methods) that have often been dismissed or are not considered by previous efforts can compete on certain types of datasets if they are tuned appropriately. Second, we find that recent methods based on matrix factorization offer a small but relatively consistent advantage over alternative methods (e.g., random-walk based methods) from a qualitative standpoint. Specifically, we find that MNMF, a community preserving embedding method, is the most competitive method for the link prediction task. While NetMF is the most competitive baseline for node classification. Third, no single method completely outperforms other embedding methods on both node classification and link prediction tasks. We also present several drill-down analysis that reveals settings under which certain algorithms perform well (e.g., the role of neighborhood context on performance) -- guiding the end-user.
Given an initial recognition model already trained on a set of base classes, the goal of this work is to develop a meta-model for few-shot learning. The meta-model, given as input some novel classes with few training examples per class, must properly adapt the existing recognition model into a new model that can correctly classify in a unified way both the novel and the base classes. To accomplish this goal it must learn to output the appropriate classification weight vectors for those two types of classes. To build our meta-model we make use of two main innovations: we propose the use of a Denoising Autoencoder network (DAE) that (during training) takes as input a set of classification weights corrupted with Gaussian noise and learns to reconstruct the target-discriminative classification weights. In this case, the injected noise on the classification weights serves the role of regularizing the weight generating meta-model. Furthermore, in order to capture the co-dependencies between different classes in a given task instance of our meta-model, we propose to implement the DAE model as a Graph Neural Network (GNN). In order to verify the efficacy of our approach, we extensively evaluate it on ImageNet based few-shot benchmarks and we report strong results that surpass prior approaches. The code and models of our paper will be published on: https://github.com/gidariss/wDAE_GNN_FewShot
Interpretable classifiers have recently witnessed an increase in attention from the data mining community because they are inherently easier to understand and explain than their more complex counterparts. Examples of interpretable classification models include decision trees, rule sets, and rule lists. Learning such models often involves optimizing hyperparameters, which typically requires substantial amounts of data and may result in relatively large models. In this paper, we consider the problem of learning compact yet accurate probabilistic rule lists for multiclass classification. Specifically, we propose a novel formalization based on probabilistic rule lists and the minimum description length (MDL) principle. This results in virtually parameter-free model selection that naturally allows to trade-off model complexity with goodness of fit, by which overfitting and the need for hyperparameter tuning are effectively avoided. Finally, we introduce the Classy algorithm, which greedily finds rule lists according to the proposed criterion. We empirically demonstrate that Classy selects small probabilistic rule lists that outperform state-of-the-art classifiers when it comes to the combination of predictive performance and interpretability. We show that Classy is insensitive to its only parameter, i.e., the candidate set, and that compression on the training set correlates with classification performance, validating our MDL-based selection criterion.
Deep learning has revolutionized computer vision utilizing the increased availability of big data and the power of parallel computational units such as graphical processing units. The vast majority of deep learning research is conducted using images as training data, however the biomedical domain is rich in physiological signals that are used for diagnosis and prediction problems. It is still an open research question how to best utilize signals to train deep neural networks. In this paper we define the term Signal2Image (S2Is) as trainable or non-trainable prefix modules that convert signals, such as Electroencephalography (EEG), to image-like representations making them suitable for training image-based deep neural networks defined as `base models'. We compare the accuracy and time performance of four S2Is (`signal as image', spectrogram, one and two layer Convolutional Neural Networks (CNNs)) combined with a set of `base models' (LeNet, AlexNet, VGGnet, ResNet, DenseNet) along with the depth-wise and 1D variations of the latter. We also provide empirical evidence that the one layer CNN S2I performs better in eleven out of fifteen tested models than non-trainable S2Is for classifying EEG signals and we present visual comparisons of the outputs of the S2Is.
In this paper, we analyze the inner product of weight vector w and data vector x in neural networks from the perspective of vector orthogonal decomposition and prove that the direction gradient of w decreases with the angle between them close to 0 or {\pi}. We propose the Projection and Rejection Product (PR Product) to make the direction gradient of w independent of the angle and consistently larger than the one in standard inner product while keeping the forward propagation identical. As a reliable substitute for standard inner product, the PR Product can be applied into many existing deep learning modules, so we develop the PR Product version of fully connected layer, convolutional layer and LSTM layer. In static image classification, the experiments on CIFAR10 and CIFAR100 datasets demonstrate that the PR Product can robustly enhance the ability of various state-of-the-art classification networks. On the task of image captioning, even without any bells and whistles, our PR Product version of captioning model can compete or outperform the state-of-the-art models on MS COCO dataset. Code has been made available at:https://github.com/wzn0828/PR_Product.
As an efficient and scalable graph neural network, GraphSAGE has enabled an inductive capability for inferring unseen nodes or graphs by aggregating subsampled local neighborhoods and by learning in a mini-batch gradient descent fashion. The neighborhood sampling used in GraphSAGE is effective in order to improve computing and memory efficiency when inferring a batch of target nodes with diverse degrees in parallel. Despite this advantage, the default uniform sampling suffers from high variance in training and inference, leading to sub-optimum accuracy. We propose a new data-driven sampling approach to reason about the real-valued importance of a neighborhood by a non-linear regressor, and to use the value as a criterion for subsampling neighborhoods. The regressor is learned using a value-based reinforcement learning. The implied importance for each combination of vertex and neighborhood is inductively extracted from the negative classification loss output of GraphSAGE. As a result, in an inductive node classification benchmark using three datasets, our method enhanced the baseline using the uniform sampling, outperforming recent variants of a graph neural network in accuracy.
Convolutional neural networks (CNNs) deliver exceptional results for computer vision, including medical image analysis. With the growing number of available architectures, picking one over another is far from obvious. Existing art suggests that, when performing transfer learning, the performance of CNN architectures on ImageNet correlates strongly with their performance on target tasks. We evaluate that claim for melanoma classification, over 9 CNNs architectures, in 5 sets of splits created on the ISIC Challenge 2017 dataset, and 3 repeated measures, resulting in 135 models. The correlations we found were, to begin with, much smaller than those reported by existing art, and disappeared altogether when we considered only the top-performing networks: uncontrolled nuisances (i.e., splits and randomness) overcome any of the analyzed factors. Whenever possible, the best approach for melanoma classification is still to create ensembles of multiple models. We compared two choices for selecting which models to ensemble: picking them at random (among a pool of high-quality ones) vs. using the validation set to determine which ones to pick first. For small ensembles, we found a slight advantage on the second approach but found that random choice was also competitive. Although our aim in this paper was not to maximize performance, we easily reached AUCs comparable to the first place on the ISIC Challenge 2017.
We present a very simple, unsupervised method for the pairwise matching of documents from heterogeneous collections. We demonstrate our method with the Concept-Project matching task, which is a binary classification task involving pairs of documents from heterogeneous collections. Although our method only employs standard resources without any domain- or task-specific modifications, it clearly outperforms the more complex system of the original authors. In addition, our method is transparent, because it provides explicit information about how a similarity score was computed, and efficient, because it is based on the aggregation of (pre-computable) word-level similarities.
Unsupervised model transfer has the potential to greatly improve the generalizability of deep models to novel domains. Yet the current literature assumes that the separation of target data into distinct domains is known as a priori. In this paper, we propose the task of Domain-Agnostic Learning (DAL): How to transfer knowledge from a labeled source domain to unlabeled data from arbitrary target domains? To tackle this problem, we devise a novel Deep Adversarial Disentangled Autoencoder (DADA) capable of disentangling domain-specific features from class identity. We demonstrate experimentally that when the target domain labels are unknown, DADA leads to state-of-the-art performance on several image classification datasets.
Discriminatively trained neural classifiers can be trusted, only when the input data comes from the training distribution (in-distribution). Therefore, detecting out-of-distribution (OOD) samples is very important to avoid classification errors. In the context of OOD detection for image classification, one of the recent approaches proposes training a classifier called "confident-classifier" by minimizing the standard cross-entropy loss on in-distribution samples and minimizing the KL divergence between the predictive distribution of OOD samples in the low-density regions of in-distribution and the uniform distribution (maximizing the entropy of the outputs). Thus, the samples could be detected as OOD if they have low confidence or high entropy. In this paper, we analyze this setting both theoretically and experimentally. We conclude that the resulting confident-classifier still yields arbitrarily high confidence for OOD samples far away from the in-distribution. We instead suggest training a classifier by adding an explicit "reject" class for OOD samples.
When translating phrases (words or group of words), human translators, consciously or not, resort to different translation processes apart from the literal translation, such as Idiom Equivalence, Generalization, Particularization, Semantic Modulation, etc. Translators and linguists (such as Vinay and Darbelnet, Newmark, etc.) have proposed several typologies to characterize the different translation processes. However, to the best of our knowledge, there has not been effort to automatically classify these fine-grained translation processes. Recently, an English-French parallel corpus of TED Talks has been manually annotated with translation process categories, along with established annotation guidelines. Based on these annotated examples, we propose an automatic classification of translation processes at subsentential level. Experimental results show that we can distinguish non-literal translation from literal translation with an accuracy of 87.09%, and 55.20% for classifying among five non-literal translation processes. This work demonstrates that it is possible to automatically classify translation processes. Even with a small amount of annotated examples, our experiments show the directions that we can follow in future work. One of our long term objectives is leveraging this automatic classification to better control paraphrase extraction from bilingual parallel corpora.
Adversarial training, in which a network is trained on adversarial examples, is one of the few defenses against adversarial attacks that withstands strong attacks. Unfortunately, the high cost of generating strong adversarial examples makes standard adversarial training impractical on large-scale problems like ImageNet. We present an algorithm that eliminates the overhead cost of generating adversarial examples by recycling the gradient information computed when updating model parameters. Our "free" adversarial training algorithm achieves comparable robustness to PGD adversarial training on the CIFAR-10 and CIFAR-100 datasets at negligible additional cost compared to natural training, and can be 7 to 30 times faster than other strong adversarial training methods. Using a single workstation with 4 P100 GPUs and 2 days of runtime, we can train a robust model for the large-scale ImageNet classification task that maintains 40% accuracy against PGD attacks. The code is available at https://github.com/ashafahi/free_adv_train.
Serial crystallography is the field of science that studies the structure and properties of crystals via diffraction patterns. In this paper, we introduce a new serial crystallography dataset comprised of real and synthetic images; the synthetic images are generated through the use of a simulator that is both scalable and accurate. The resulting dataset is called DiffraNet, and it is composed of 25,457 512x512 grayscale labeled images. We explore several computer vision approaches for classification on DiffraNet such as standard feature extraction algorithms associated with Random Forests and Support Vector Machines but also an end-to-end CNN topology dubbed DeepFreak tailored to work on this new dataset. All implementations are publicly available and have been fine-tuned using off-the-shelf AutoML optimization tools for a fair comparison. Our best model achieves 98.5% accuracy on synthetic images and 94.51% accuracy on real images. We believe that the DiffraNet dataset and its classification methods will have in the long term a positive impact in accelerating discoveries in many disciplines, including chemistry, geology, biology, materials science, metallurgy, and physics.
Graph convolutional neural networks (Graph-CNNs) extend traditional CNNs to handle data that is supported on a graph. Major challenges when working with data on graphs are that the support set (the vertices of the graph) do not typically have a natural ordering, and in general, the topology of the graph is not regular (i.e., vertices do not all have the same number of neighbors). Thus, Graph-CNNs have huge potential to deal with 3D point cloud data which has been obtained from sampling a manifold. In this paper, we develop a Graph-CNN for classifying 3D point cloud data, called PointGCN. The architecture combines localized graph convolutions with two types of graph downsampling operations (also known as pooling). By the effective exploration of the point cloud local structure using the Graph-CNN, the proposed architecture achieves competitive performance on the 3D object classification benchmark ModelNet, and our architecture is more stable than competing schemes.