This paper presents a Depthwise Disout Convolutional Neural Network (DD-CNN) for the detection and classification of urban acoustic scenes. Specifically, we use log-mel as feature representations of acoustic signals for the inputs of our network. In the proposed DD-CNN, depthwise separable convolution is used to reduce the network complexity. Besides, SpecAugment and Disout are used for further performance boosting. Experimental results demonstrate that our DD-CNN can learn discriminative acoustic characteristics from audio fragments and effectively reduce the network complexity. Our DD-CNN was used for the low-complexity acoustic scene classification task of the DCASE2020 Challenge, which achieves 92.04% accuracy on the validation set.
Convolutional Neural Networks (CNNs) have been successfully used in various Music Information Retrieval (MIR) tasks, both as end-to-end models and as feature extractors for more complex systems. However, the MIR field is still dominated by the classical VGG-based CNN architecture variants, often in combination with more complex modules such as attention, and/or techniques such as pre-training on large datasets. Deeper models such as ResNet -- which surpassed VGG by a large margin in other domains -- are rarely used in MIR. One of the main reasons for this, as we will show, is the lack of generalization of deeper CNNs in the music domain. In this paper, we present a principled way to make deep architectures like ResNet competitive for music-related tasks, based on well-designed regularization strategies. In particular, we analyze the recently introduced Receptive-Field Regularization and Shake-Shake, and show that they significantly improve the generalization of deep CNNs on music-related tasks, and that the resulting deep CNNs can outperform current more complex models such as CNNs augmented with pre-training and attention. We demonstrate this on two different MIR tasks and two corresponding datasets, thus offering our deep regularized CNNs as a new baseline for these datasets, which can also be used as a feature-extracting module in future, more complex approaches.
Graph neural networks (GNNs) achieve remarkable success in graph-based semi-supervised node classification, leveraging the information from neighboring nodes to improve the representation learning of target node. The success of GNNs at node classification depends on the assumption that connected nodes tend to have the same label. However, such an assumption does not always work, limiting the performance of GNNs at node classification. In this paper, we propose label-consistency based graph neural network(LC-GNN), leveraging node pairs unconnected but with the same labels to enlarge the receptive field of nodes in GNNs. Experiments on benchmark datasets demonstrate the proposed LC-GNN outperforms traditional GNNs in graph-based semi-supervised node classification.We further show the superiority of LC-GNN in sparse scenarios with only a handful of labeled nodes.
The popular softmax loss and its recent extensions have achieved great success in the deep learning-based image classification. However, the data for training image classifiers usually has different quality. Ignoring such problem, the correct classification of low quality data is hard to be solved. In this paper, we discover the positive correlation between the feature norm of an image and its quality through careful experiments on various applications and various deep neural networks. Based on this finding, we propose a contraction mapping function to compress the range of feature norms of training images according to their quality and embed this contraction mapping function into softmax loss or its extensions to produce novel learning objectives. The experiments on various classification applications, including handwritten digit recognition, lung nodule classification, face verification and face recognition, demonstrate that the proposed approach is promising to effectively deal with the problem of learning on the data with different quality and leads to the significant and stable improvements in the classification accuracy.
In the image classification task, the most common approach is to resize all images in a dataset to a unique shape, while reducing their precision to a size which facilitates experimentation at scale. This practice has benefits from a computational perspective, but it entails negative side-effects on performance due to loss of information and image deformation. In this work we introduce the MAMe dataset, an image classification dataset with remarkable high resolution and variable shape properties. The goal of MAMe is to provide a tool for studying the impact of such properties in image classification, while motivating research in the field. The MAMe dataset contains thousands of artworks from three different museums, and proposes a classification task consisting on differentiating between 29 mediums (i.e. materials and techniques) supervised by art experts. After reviewing the singularity of MAMe in the context of current image classification tasks, a thorough description of the task is provided, together with dataset statistics. Experiments are conducted to evaluate the impact of using high resolution images, variable shape inputs and both properties at the same time. Results illustrate the positive impact in performance when using high resolution images, while highlighting the lack of solutions to exploit variable shapes. An additional experiment exposes the distinctiveness between the MAMe dataset and the prototypical ImageNet dataset. Finally, the baselines are inspected using explainability methods and expert knowledge, to gain insights on the challenges that remain ahead.
In this paper, a novel multi-view methodology for graph-based neural networks is proposed. A systematic and methodological adaptation of the key concepts of classical deep learning methods such as convolution, pooling and multi-view architectures is developed for the context of non-Euclidean manifolds. The aim of the proposed work is to present a novel multi-view graph convolution layer, as well as a new view pooling layer making use of: a) a new hybrid Laplacian that is adjusted based on feature distance metric learning, b) multiple trainable representations of a feature matrix of a graph, using trainable distance matrices, adapting the notion of views to graphs and c) a multi-view graph aggregation scheme called graph view pooling, in order to synthesise information from the multiple generated views. The aforementioned layers are used in an end-to-end graph neural network architecture for graph classification and show competitive results to other state-of-the-art methods.
We consider Multiclass and Multilabel classification with extremely large number of classes, of which only few are labeled to each instance. In such setting, standard methods that have training, prediction cost linear to the number of classes become intractable. State-of-the-art methods thus aim to reduce the complexity by exploiting correlation between labels under assumption that the similarity between labels can be captured by structures such as low-rank matrix or balanced tree. However, as the diversity of labels increases in the feature space, structural assumption can be easily violated, which leads to degrade in the testing performance. In this work, we show that a margin-maximizing loss with l1 penalty, in case of Extreme Classification, yields extremely sparse solution both in primal and in dual without sacrificing the expressive power of predictor. We thus propose a Fully-Corrective Block-Coordinate Frank-Wolfe (FC-BCFW) algorithm that exploits both primal and dual sparsity to achieve a complexity sublinear to the number of primal and dual variables. A bi-stochastic search method is proposed to further improve the efficiency. In our experiments on both Multiclass and Multilabel problems, the proposed method achieves significant higher accuracy than existing approaches of Extreme Classification with very competitive training and prediction time.
Decision tree learning is a widely used approach in machine learning, favoured in applications that require concise and interpretable models. Heuristic methods are traditionally used to quickly produce models with reasonably high accuracy. A commonly criticised point, however, is that the resulting trees may not necessarily be the best representation of the data in terms of accuracy and size. In recent years, this motivated the development of optimal classification tree algorithms that globally optimise the decision tree in contrast to heuristic methods that perform a sequence of locally optimal decisions. We follow this line of work and provide a novel algorithm for learning optimal classification trees based on dynamic programming and search. Our algorithm supports constraints on the depth of the tree and number of nodes. The success of our approach is attributed to a series of specialised techniques that exploit properties unique to classification trees. Whereas algorithms for optimal classification trees have traditionally been plagued by high runtimes and limited scalability, we show in a detailed experimental study that our approach uses only a fraction of the time required by the state-of-the-art and can handle datasets with tens of thousands of instances, providing several orders of magnitude improvements and notably contributing towards the practical realisation of optimal decision trees.
We present Gaptron, a randomized first-order algorithm for online multiclass classification. In the full information setting we show expected mistake bounds with respect to the logistic loss, hinge loss, and the smooth hinge loss with constant regret, where the expectation is with respect to the learner's randomness. In the bandit classification setting we show that Gaptron is the first linear time algorithm with $O(K\sqrt{T})$ expected regret, where $K$ is the number of classes. Additionally, the expected mistake bound of Gaptron does not depend on the dimension of the feature vector, contrary to previous algorithms with $O(K\sqrt{T})$ regret in the bandit classification setting. We present a new proof technique that exploits the gap between the zero-one loss and surrogate losses rather than exploiting properties such as exp-concavity or mixability, which are traditionally used to prove logarithmic or constant regret bounds.
Classification is a well-studied machine learning task which concerns the assignment of instances to a set of outcomes. Classification models support the optimization of managerial decision-making across a variety of operational business processes. For instance, customer churn prediction models are adopted to increase the efficiency of retention campaigns by optimizing the selection of customers that are to be targeted. Cost-sensitive and causal classification methods have independently been proposed to improve the performance of classification models. The former considers the benefits and costs of correct and incorrect classifications, such as the benefit of a retained customer, whereas the latter estimates the causal effect of an action, such as a retention campaign, on the outcome of interest. This study integrates cost-sensitive and causal classification by elaborating a unifying evaluation framework. The framework encompasses a range of existing and novel performance measures for evaluating both causal and conventional classification models in a cost-sensitive as well as a cost-insensitive manner. We proof that conventional classification is a specific case of causal classification in terms of a range of performance measures when the number of actions is equal to one. The framework is shown to instantiate to application-specific cost-sensitive performance measures that have been recently proposed for evaluating customer retention and response uplift models, and allows to maximize profitability when adopting a causal classification model for optimizing decision-making. The proposed framework paves the way toward the development of cost-sensitive causal learning methods and opens a range of opportunities for improving data-driven business decision-making.
Most of the saliency methods are evaluated on their ability to generate saliency maps, and not on their functionality in a complete vision pipeline, like for instance, image classification. In the current paper, we propose an approach which does not require explicit saliency maps to improve image classification, but they are learned implicitely, during the training of an end-to-end image classification task. We show that our approach obtains similar results as the case when the saliency maps are provided explicitely. Combining RGB data with saliency maps represents a significant advantage for object recognition, especially for the case when training data is limited. We validate our method on several datasets for fine-grained classification tasks (Flowers, Birds and Cars). In addition, we show that our saliency estimation method, which is trained without any saliency groundtruth data, obtains competitive results on real image saliency benchmark (Toronto), and outperforms deep saliency models with synthetic images (SID4VAM).
In this paper, we present our approach for sentiment classification on Spanish-English code-mixed social media data in the SemEval-2020 Task 9. We investigate performance of various pre-trained Transformer models by using different fine-tuning strategies. We explore both monolingual and multilingual models with the standard fine-tuning method. Additionally, we propose a custom model that we fine-tune in two steps: once with a language modeling objective, and once with a task-specific objective. Although two-step fine-tuning improves sentiment classification performance over the base model, the large multilingual XLM-RoBERTa model achieves best weighted F1-score with 0.537 on development data and 0.739 on test data. With this score, our team jupitter placed tenth overall in the competition.
It has been proved that deep neural networks are facing a new threat called backdoor attacks, where the adversary can inject backdoors into the neural network model through poisoning the training dataset. When the input containing some special pattern called the backdoor trigger, the model with backdoor will carry out malicious task such as misclassification specified by adversaries. In text classification systems, backdoors inserted in the models can cause spam or malicious speech to escape detection. Previous work mainly focused on the defense of backdoor attacks in computer vision, little attention has been paid to defense method for RNN backdoor attacks regarding text classification. In this paper, through analyzing the changes in inner LSTM neurons, we proposed a defense method called Backdoor Keyword Identification (BKI) to mitigate backdoor attacks which the adversary performs against LSTM-based text classification by data poisoning. This method can identify and exclude poisoning samples crafted to insert backdoor into the model from training data without a verified and trusted dataset. We evaluate our method on four different text classification datset: IMDB, DBpedia ontology, 20 newsgroups and Reuters-21578 dataset. It all achieves good performance regardless of the trigger sentences.
Red blood cells are highly deformable and present in various shapes. In blood cell disorders, only a subset of all cells is morphologically altered and relevant for the diagnosis. However, manually labeling of all cells is laborious, complicated and introduces inter-expert variability. We propose an attention based multiple instance learning method to classify blood samples of patients suffering from blood cell disorders. Cells are detected using an R-CNN architecture. With the features extracted for each cell, a multiple instance learning method classifies patient samples into one out of four blood cell disorders. The attention mechanism provides a measure of the contribution of each cell to the overall classification and significantly improves the network's classification accuracy as well as its interpretability for the medical expert.
Light Detection and Ranging (LiDAR) data are widely used for high-resolution land cover mapping, which can provide very valuable information about the height of the surveyed area for the discrim- ination of classes. In order to utilize the advantages of deep models for the classification of LiDAR-derived features, a new classification algorithm combined Octave Convolution (OctConv) with Capsule Network (CapsNet), is proposed here to hierarchically extract robust and discriminant features of the input data, called as OctConv-CapsNet. In the proposed approach, CapsNet captures the spatial information of the data, and OctConv processes separately for high- and low-frequency feature. OctConv is embedded in the primary capsule layer of CapsNet so that the proposed approach can make the most of both the spatial information and the high- and low-frequency information simultaneously. The proposed framework performs experiments on two LiDAR-DSM datasets (i.e. Bayview Park and Recology datasets). The results show that, compared with the traditional deep convolution model, OctConv-CapsNet can improve the classification accuracy of LiDAR-DSM data, and when the number of training samples of the experiment is 800, the classification accuracies reached 96.12% and 96.79% on Bayview Park and Recology datasets, respectively.
Convolutional operations have two limitations: (1) do not explicitly model where to focus as the same filter is applied to all the positions, and (2) are unsuitable for modeling long-range dependencies as they only operate on a small neighborhood. While both limitations can be alleviated by attention operations, many design choices remain to be determined to use attention, especially when applying attention to videos. Towards a principled way of applying attention to videos, we address the task of spatiotemporal attention cell search. We propose a novel search space for spatiotemporal attention cells, which allows the search algorithm to flexibly explore various design choices in the cell. The discovered attention cells can be seamlessly inserted into existing backbone networks, e.g., I3D or S3D, and improve video classification accuracy by more than 2% on both Kinetics-600 and MiT datasets. The discovered attention cells outperform non-local blocks on both datasets, and demonstrate strong generalization across different modalities, backbones, and datasets. Inserting our attention cells into I3D-R50 yields state-of-the-art performance on both datasets.
The effectiveness of Convolutional Neural Networks (CNNs)in classifying image data has been thoroughly demonstrated. In order to explain the classification to humans, methods for visualizing classification evidence have been developed in recent years. These explanations reveal that sometimes images are classified correctly, but for the wrong reasons,i.e., based on incidental evidence. Of course, it is desirable that images are classified correctly for the right reasons, i.e., based on the actual evidence. To this end, we propose a new explanation quality metric to measure object aligned explanation in image classification which we refer to as theObAlExmetric. Using object detection approaches, explanation approaches, and ObAlEx, we quantify the focus of CNNs on the actual evidence. Moreover, we show that additional training of the CNNs can improve the focus of CNNs without decreasing their accuracy.
We present a new embedding-based framework for zero-shot learning (ZSL). Most embedding-based methods aim to learn the correspondence between an image classifier (visual representation) and its class prototype (semantic representation) for each class. Motivated by the binary relevance method for multi-label classification, we propose to inversely learn the mapping between an image and a semantic classifier. Given an input image, the proposed Image-Guided Semantic Classification (IGSC) method creates a label classifier, being applied to all label embeddings to determine whether a label belongs to the input image. Therefore, semantic classifiers are image-adaptive and are generated during inference. IGSC is conceptually simple and can be realized by a slight enhancement of an existing deep architecture for classification; yet it is effective and outperforms state-of-the-art embedding-based generalized ZSL approaches on standard benchmarks.
Cervical cancer is one of the deadliest cancers affecting women globally. Cervical intraepithelial neoplasia (CIN) assessment using histopathological examination of cervical biopsy slides is subject to interobserver variability. Automated processing of digitized histopathology slides has the potential for more accurate classification for CIN grades from normal to increasing grades of pre-malignancy: CIN1, CIN2 and CIN3. Cervix disease is generally understood to progress from the bottom (basement membrane) to the top of the epithelium. To model this relationship of disease severity to spatial distribution of abnormalities, we propose a network pipeline, DeepCIN, to analyze high-resolution epithelium images (manually extracted from whole-slide images) hierarchically by focusing on localized vertical regions and fusing this local information for determining Normal/CIN classification. The pipeline contains two classifier networks: 1) a cross-sectional, vertical segment-level sequence generator (two-stage encoder model) is trained using weak supervision to generate feature sequences from the vertical segments to preserve the bottom-to-top feature relationships in the epithelium image data; 2) an attention-based fusion network image-level classifier predicting the final CIN grade by merging vertical segment sequences. The model produces the CIN classification results and also determines the vertical segment contributions to CIN grade prediction. Experiments show that DeepCIN achieves pathologist-level CIN classification accuracy.
Recently, deep learning has been adopted to the glaucoma classification task with performance comparable to that of human experts. However, a well trained deep learning model demands a large quantity of properly labeled data, which is relatively expensive since the accurate labeling of glaucoma requires years of specialist training. In order to alleviate this problem, we propose a glaucoma classification framework which takes advantage of not only the properly labeled images, but also undiagnosed images without glaucoma labels. To be more specific, the proposed framework is adapted from the teacher-student-learning paradigm. The teacher model encodes the wrapped information of undiagnosed images to a latent feature space, meanwhile the student model learns from the teacher through knowledge transfer to improve the glaucoma classification. For the model training procedure, we propose a novel training strategy that simulates the real-world teaching practice named as 'Learning To Teach with Knowledge Transfer (L2T-KT)', and establish a 'Quiz Pool' as the teacher's optimization target. Experiments show that the proposed framework is able to utilize the undiagnosed data effectively to improve the glaucoma prediction performance.