The natural world often follows a long-tailed data distribution where only a few classes account for most of the examples. This long-tail causes classifiers to overfit to the majority class. To mitigate this, prior solutions commonly adopt class rebalancing strategies such as data resampling and loss reshaping. However, by treating each example within a class equally, these methods fail to account for the important notion of example hardness, i.e., within each class some examples are easier to classify than others. To incorporate this notion of hardness into the learning process, we propose the EarLy-exiting Framework(ELF). During training, ELF learns to early-exit easy examples through auxiliary branches attached to a backbone network. This offers a dual benefit-(1) the neural network increasingly focuses on hard examples, since they contribute more to the overall network loss; and (2) it frees up additional model capacity to distinguish difficult examples. Experimental results on two large-scale datasets, ImageNet LT and iNaturalist'18, demonstrate that ELF can improve state-of-the-art accuracy by more than 3 percent. This comes with the additional benefit of reducing up to 20 percent of inference time FLOPS. ELF is complementary to prior work and can naturally integrate with a variety of existing methods to tackle the challenge of long-tailed distributions.
While a variety of ensemble methods for multilabel classification have been proposed in the literature, the question of how to aggregate the predictions of the individual members of the ensemble has received little attention so far. In this paper, we introduce a formal framework of ensemble multilabel classification, in which we distinguish two principal approaches: "predict then combine" (PTC), where the ensemble members first make loss minimizing predictions which are subsequently combined, and "combine then predict" (CTP), which first aggregates information such as marginal label probabilities from the individual ensemble members, and then derives a prediction from this aggregation. While both approaches generalize voting techniques commonly used for multilabel ensembles, they allow to explicitly take the target performance measure into account. Therefore, concrete instantiations of CTP and PTC can be tailored to concrete loss functions. Experimentally, we show that standard voting techniques are indeed outperformed by suitable instantiations of CTP and PTC, and provide some evidence that CTP performs well for decomposable loss functions, whereas PTC is the better choice for non-decomposable losses.
One intriguing property of deep neural networks (DNNs) is their inherent vulnerability to backdoor attacks -- a trojan model responds to trigger-embedded inputs in a highly predictable manner while functioning normally otherwise. Despite the plethora of prior work on DNNs for continuous data (e.g., images), the vulnerability of graph neural networks (GNNs) for discrete-structured data (e.g., graphs) is largely unexplored, which is highly concerning given their increasing use in security-sensitive domains. To bridge this gap, we present GTA, the first backdoor attack on GNNs. Compared with prior work, GTA departs in significant ways: graph-oriented -- it defines triggers as specific subgraphs, including both topological structures and descriptive features, entailing a large design spectrum for the adversary; input-tailored -- it dynamically adapts triggers to individual graphs, thereby optimizing both attack effectiveness and evasiveness; downstream model-agnostic -- it can be readily launched without knowledge regarding downstream models or fine-tuning strategies; and attack-extensible -- it can be instantiated for both transductive (e.g., node classification) and inductive (e.g., graph classification) tasks, constituting severe threats for a range of security-critical applications. Through extensive evaluation using benchmark datasets and state-of-the-art models, we demonstrate the effectiveness of GTA. We further provide analytical justification for its effectiveness and discuss potential countermeasures, pointing to several promising research directions.
Most supervised learning models are trained for full automation. However, their predictions are sometimes worse than those by human experts on some specific instances. Motivated by this empirical observation, our goal is to design classifiers that are optimized to operate under different automation levels. More specifically, we focus on convex margin-based classifiers and first show that the problem is NP-hard. Then, we further show that, for support vector machines, the corresponding objective function can be expressed as the difference of two functions f = g - c, where g is monotone, non-negative and {\gamma}-weakly submodular, and c is non-negative and modular. This representation allows a recently introduced deterministic greedy algorithm, as well as a more efficient randomized variant of the algorithm, to enjoy approximation guarantees at solving the problem. Experiments on synthetic and real-world data from several applications in medical diagnosis illustrate our theoretical findings and demonstrate that, under human assistance, supervised learning models trained to operate under different automation levels can outperform those trained for full automation as well as humans operating alone.
Pixel based algorithms including back propagation neural networks (NN) and support vector machines (SVM) have been widely used for remotely sensed image classifications. Within last few years, deep learning based image classifier like convolution neural networks (2D-CNN) are becoming popular alternatives to these classifiers. In this paper, we compare performance of patch based SVM and NN with that of a deep learning algorithms comprising of 2D-CNN and fully connected layers. Similar to CNN which utilise image patches to derive features for further classification, we propose to use patches as an input in place of individual pixel with both SVM and NN classifiers. Two datasets, one multispectral and other hyperspectral data was used to compare the performance of different classifiers. Results with both datasets suggest the effectiveness of patch based SVM and NN classifiers in comparison to state of art 2D-CNN classifier.
This work presents a Convolutional Neural Network (CNN) for the prediction of next-day stock fluctuations using company-specific news headlines. Experiments to evaluate model performance using various configurations of word-embeddings and convolutional filter widths are reported. The total number of convolutional filters used is far fewer than is common, reducing the dimensionality of the task without loss of accuracy. Furthermore, multiple hidden layers with decreasing dimensionality are employed. A classification accuracy of 61.7\% is achieved using pre-learned embeddings, that are fine-tuned during training to represent the specific context of this task. Multiple filter widths are also implemented to detect different length phrases that are key for classification. Trading simulations are conducted using the presented classification results. Initial investments are more than tripled over a 838 day testing period using the optimal classification configuration and a simple trading strategy. Two novel methods are presented to reduce the risk of the trading simulations. Adjustment of the sigmoid class threshold and re-labelling headlines using multiple classes form the basis of these methods. A combination of these approaches is found to more than double the Average Trade Profit (ATP) achieved during baseline simulations.
Due to the superiority and noteworthy progress of Quantum Computing (QC) in a lot of applications such as cryptography, chemistry, Big data, machine learning, optimization, Internet of Things (IoT), Blockchain, communication, and many more. Fully towards to combine classical machine learning (ML) with Quantum Information Processing (QIP) to build a new field in the quantum world is called Quantum Machine Learning (QML) to solve and improve problems that displayed in classical machine learning (e.g. time and energy consumption, kernel estimation). The aim of this paper presents and summarizes a comprehensive survey of the state-of-the-art advances in Quantum Machine Learning (QML). Especially, recent QML classification works. Also, we cover about 30 publications that are published lately in Quantum Machine Learning (QML). we propose a classification scheme in the quantum world and discuss encoding methods for mapping classical data to quantum data. Then, we provide quantum subroutines and some methods of Quantum Computing (QC) in improving performance and speed up of classical Machine Learning (ML). And also some of QML applications in various fields, challenges, and future vision will be presented.
In this paper we present Spektral, an open-source Python library for building graph neural networks with TensorFlow and the Keras application programming interface. Spektral implements a large set of methods for deep learning on graphs, including message-passing and pooling operators, as well as utilities for processing graphs and loading popular benchmark datasets. The purpose of this library is to provide the essential building blocks for creating graph neural networks, focusing on the guiding principles of user-friendliness and quick prototyping on which Keras is based. Spektral is, therefore, suitable for absolute beginners and expert deep learning practitioners alike. In this work, we present an overview of Spektral's features and report the performance of the methods implemented by the library in scenarios of node classification, graph classification, and graph regression.
Despite several signs of progress have been made recently, limited research has been conducted for an inductive setting where embeddings are required for newly unseen nodes -- a setting encountered commonly in practical applications of deep learning for graph networks. This significantly affects the performances of downstream tasks such as node classification, link prediction or community extraction. To this end, we propose SANNE -- a novel unsupervised embedding model -- whose central idea is to employ a transformer self-attention network to iteratively aggregate vector representations of nodes in random walks. Our SANNE aims to produce plausible embeddings not only for present nodes, but also for newly unseen nodes. Experimental results show that the proposed SANNE obtains state-of-the-art results for the node classification task on well-known benchmark datasets.
Texture is a fundamental characteristic of many types of images, especially those with significant rotation, scale illumination, and viewpoint change. Texture image classification is one of the challenging problems that have various applications such as remote sensing, material recognition, and computer-aided medical diagnosis, etc. Various Computer vision techniques have been used. More recently, Deep learning architectures demonstrated impressive results. This paper aims to investigate combining two feature extraction methods: Handcrafted-based and CNN-based in a two-stream neural network architecture. We believe that Statistical features could enhance the performance of the CNN architecture, especially in the case of small datasets. To test our approach we used two challenging datasets, the Describable Textures Dataset (DTD) and Flicker Material Database (FMD). Results showed that our two-stream neural network which has an image as a first stream and a statistical feature vector as a second stream achieve better results than a Convolutional neural network achieved with just the RGB image as input. The Xception network [9] combined with SIFT-FV demonstrated an accuracy superiority for both datasets.
The past years have witnessed great progress on remote sensing (RS) image interpretation and its wide applications. With RS images becoming more accessible than ever before, there is an increasing demand for the automatic interpretation of these images. In this context, the benchmark datasets serve as essential prerequisites for developing and testing intelligent interpretation algorithms. After reviewing existing benchmark datasets in the research community of RS image interpretation, this article discusses the problem of how to efficiently prepare a suitable benchmark dataset for RS image interpretation. Specifically, we first analyze the current challenges of developing intelligent algorithms for RS image interpretation with bibliometric investigations. We then present the general guidances on creating benchmark datasets in efficient manners. Following the presented guidances, we also provide an example on building RS image dataset, i.e., Million-AID, a new large-scale benchmark dataset containing a million instances for RS image scene classification. Several challenges and perspectives in RS image annotation are finally discussed to facilitate the research in benchmark dataset construction. We do hope this paper will provide the RS community an overall perspective on constructing large-scale and practical image datasets for further research, especially data-driven ones.
Deep Convolutional Neural Networks are able to identify complex patterns and perform tasks with super-human capabilities. However, besides the exceptional results, they are not completely understood and it is still impractical to hand-engineer similar solutions. In this work, an image classification Convolutional Neural Network and its building blocks are described from a frequency domain perspective. Some network layers have established counterparts in the frequency domain like the convolutional and pooling layers. We propose the 2SReLU layer, a novel non-linear activation function that preserves high frequency components in deep networks. It is demonstrated that in the frequency domain it is possible to achieve competitive results without using the computationally costly convolution operation. A source code implementation in PyTorch is provided at: https://gitlab.com/thomio/2srelu
In this paper we develop methods to solve two problems related to time series (TS) analysis using quantum computing: reconstruction and classification. We formulate the task of reconstructing a given TS from a training set of data as an unconstrained binary optimization (QUBO) problem, which can be solved by both quantum annealers and gate-model quantum processors. We accomplish this by discretizing the TS and converting the reconstruction to a set cover problem, allowing us to perform a one-versus-all method of reconstruction. Using the solution to the reconstruction problem, we show how to extend this method to perform semi-supervised classification of TS data. We present results indicating our method is competitive with current semi- and unsupervised classification techniques, but using less data than classical techniques.
Medical image analysis typically includes several tasks such as enhancement, segmentation, and classification. Traditionally, these tasks are implemented using separate deep learning models for separate tasks, which is not efficient because it involves unnecessary training repetitions, demands greater computational resources, and requires a relatively large amount of labeled data. In this paper, we propose a multi-task training approach for medical image analysis, where individual tasks are fine-tuned simultaneously through relevant knowledge transfer using a unified modality-specific feature representation (UMS-Rep). We explore different fine-tuning strategies to demonstrate the impact of the strategy on the performance of target medical image tasks. We experiment with different visual tasks (e.g., image denoising, segmentation, and classification) to highlight the advantages offered with our approach for two imaging modalities, chest X-ray and Doppler echocardiography. Our results demonstrate that the proposed approach reduces the overall demand for computational resources and improves target task generalization and performance. Further, our results prove that the performance of target tasks in medical images is highly influenced by the utilized fine-tuning strategy.
Neural networks have been shown to perform incredibly well in classification tasks over structured high-dimensional datasets. However, the learning dynamics of such networks is still poorly understood. In this paper we study in detail the training dynamics of a simple type of neural network: a single hidden layer trained to perform a classification task. We show that in a suitable mean-field limit this case maps to a single-node learning problem with a time-dependent dataset determined self-consistently from the average nodes population. We specialize our theory to the prototypical case of a linearly separable dataset and a linear hinge loss, for which the dynamics can be explicitly solved. This allow us to address in a simple setting several phenomena appearing in modern networks such as slowing down of training dynamics, crossover between rich and lazy learning, and overfitting. Finally, we asses the limitations of mean-field theory by studying the case of large but finite number of nodes and of training samples.
In this work, we propose the first backdoor attack to graph neural networks (GNN). Specifically, we propose a \emph{subgraph based backdoor attack} to GNN for graph classification. In our backdoor attack, a GNN classifier predicts an attacker-chosen target label for a testing graph once a predefined subgraph is injected to the testing graph. Our empirical results on three real-world graph datasets show that our backdoor attacks are effective with a small impact on a GNN's prediction accuracy for clean testing graphs. Moreover, we generalize a randomized smoothing based certified defense to defend against our backdoor attacks. Our empirical results show that the defense is effective in some cases but ineffective in other cases, highlighting the needs of new defenses for our backdoor attacks.
Exploratory landscape analysis (ELA) supports supervised learning approaches for automated algorithm selection and configuration by providing sets of features that quantify the most relevant characteristics of the optimization problem at hand. In black-box optimization, where an explicit problem representation is not available, the feature values need to be approximated from a small number of sample points. In practice, uniformly sampled random point sets and Latin hypercube constructions are commonly used sampling strategies. In this work, we analyze how the sampling method and the sample size influence the quality of the feature value approximations and how this quality impacts the accuracy of a standard classification task. While, not unexpectedly, increasing the number of sample points gives more robust estimates for the feature values, to our surprise we find that the feature value approximations for different sampling strategies do not converge to the same value. This implies that approximated feature values cannot be interpreted independently of the underlying sampling strategy. As our classification experiments show, this also implies that the feature approximations used for training a classifier must stem from the same sampling strategy as those used for the actual classification tasks. As a side result we show that classifiers trained with feature values approximated by Sobol' sequences achieve higher accuracy than any of the standard sampling techniques. This may indicate improvement potential for ELA-trained machine learning models.
Classifiers are often tested on relatively small data sets, which should lead to uncertain performance metrics. Nevertheless, these metrics are usually taken at face value. We present an approach to quantify the uncertainty of classification performance metrics, based on a probability model of the confusion matrix. Application of our approach to classifiers from the scientific literature and a classification competition shows that uncertainties can be surprisingly large and limit performance evaluation. In fact, some published classifiers are likely to be misleading. The application of our approach is simple and requires only the confusion matrix. It is agnostic of the underlying classifier. Our method can also be used for the estimation of sample sizes that achieve a desired precision of a performance metric.
Instance level detection of thoracic diseases or abnormalities are crucial for automatic diagnosis in chest X-ray images. Most existing works on chest X-rays focus on disease classification and weakly supervised localization. In order to push forward the research on disease classification and localization on chest X-rays. We provide a new benchmark called ChestX-Det10, including box-level annotations of 10 categories of disease/abnormality of $\sim$ 3,500 images. The annotations are located at https://github.com/Deepwise-AILab/ChestX-Det10-Dataset.
There are several ways to categorise reinforcement learning (RL) algorithms, such as either model-based or model-free, policy-based or planning-based, on-policy or off-policy, and online or offline. Broad classification schemes such as these help provide a unified perspective on disparate techniques and can contextualise and guide the development of new algorithms. In this paper, we utilise the control as inference framework to outline a novel classification scheme based on amortised and iterative inference. We demonstrate that a wide range of algorithms can be classified in this manner providing a fresh perspective and highlighting a range of existing similarities. Moreover, we show that taking this perspective allows us to identify parts of the algorithmic design space which have been relatively unexplored, suggesting new routes to innovative RL algorithms.