Despite the wide empirical success of modern machine learning algorithms and models in a multitude of applications, they are known to be highly susceptible to seemingly small indiscernible perturbations to the input data known as \emph{adversarial attacks}. A variety of recent adversarial training procedures have been proposed to remedy this issue. Despite the success of such procedures at increasing accuracy on adversarially perturbed inputs or \emph{robust accuracy}, these techniques often reduce accuracy on natural unperturbed inputs or \emph{standard accuracy}. Complicating matters further, the effect and trend of adversarial training procedures on standard and robust accuracy is rather counter intuitive and radically dependent on a variety of factors including the perceived form of the perturbation during training, size/quality of data, model overparameterization, etc. In this paper we focus on binary classification problems where the data is generated according to the mixture of two Gaussians with general anisotropic covariance matrices and derive a precise characterization of the standard and robust accuracy for a class of minimax adversarially trained models. We consider a general norm-based adversarial model, where the adversary can add perturbations of bounded $\ell_p$ norm to each input data, for an arbitrary $p\ge 1$. Our comprehensive analysis allows us to theoretically explain several intriguing empirical phenomena and provide a precise understanding of the role of different problem parameters on standard and robust accuracies.
The goal of classification with rejection is to avoid risky misclassification in error-critical applications such as medical diagnosis and product inspection. In this paper, based on the relationship between classification with rejection and cost-sensitive classification, we propose a novel method of classification with rejection by learning an ensemble of cost-sensitive classifiers, which satisfies all the following properties: (i) it can avoid estimating class-posterior probabilities, resulting in improved classification accuracy, (ii) it allows a flexible choice of losses including non-convex ones, (iii) it does not require complicated modifications when using different losses, (iv) it is applicable to both binary and multiclass cases, and (v) it is theoretically justifiable for any classification-calibrated loss. Experimental results demonstrate the usefulness of our proposed approach in clean-labeled, noisy-labeled, and positive-unlabeled classification.
The advent of Deep Learning models like VGG-16 and Resnet-50 has considerably revolutionized the field of image classification, and by using these Convolutional Neural Networks (CNN) architectures, one can get a high classification accuracy on a wide variety of image datasets. However, these Deep Learning models have a very high computational complexity and so incur a high computational cost of running these algorithms as well as make it hard to interpret the results. In this paper, we present a shallow CNN architecture which gives the same classification accuracy as the VGG-16 and Resnet-50 models for thin blood smear RBC slide images for detection of malaria, while decreasing the computational run time by an order of magnitude. This can offer a significant advantage for commercial deployment of these algorithms, especially in poorer countries in Africa and some parts of the Indian subcontinent, where the menace of malaria is quite severe.
The mood of a song is a highly relevant feature for exploration and recommendation in large collections of music. These collections tend to require automatic methods for predicting such moods. In this work, we show that listening-based features outperform content-based ones when classifying moods: embeddings obtained through matrix factorization of listening data appear to be more informative of a track mood than embeddings based on its audio content. To demonstrate this, we compile a subset of the Million Song Dataset, totalling 67k tracks, with expert annotations of 188 different moods collected from AllMusic. Our results on this novel dataset not only expose the limitations of current audio-based models, but also aim to foster further reproducible research on this timely topic.
Urban sound classification has been achieving remarkable progress and is still an active research area in audio pattern recognition. In particular, it allows to monitor the noise pollution, which becomes a growing concern for large cities. The contribution of this paper is two-fold. First, we present our DCASE 2020 task 5 winning solution which aims at helping the monitoring of urban noise pollution. It achieves a macro-AUPRC of 0.82 / 0.62 for the coarse / fine classification on validation set. Moreover, it reaches accuracies of 89.7% and 85.41% respectively on ESC-50 and US8k datasets. Second, it is not easy to find a fair comparison and to reproduce the performance of existing models. Sometimes authors copy-pasting the results of the original papers which is not helping reproducibility. As a result, we provide a fair comparison by using the same input representation, metrics and optimizer to assess performances. We preserve data augmentation used by the original papers. We hope this framework could help evaluate new architectures in this field. For better reproducibility, the code is available on our GitHub repository.
The rapid development in data collecting devices and computation platforms produces an emerging number of agents, each equipped with a unique data modality over a particular population of subjects. While the predictive performance of an agent may be enhanced by transmitting other data to it, this is often unrealistic due to intractable transmission costs and security concerns. While the predictive performance of an agent may be enhanced by transmitting other data to it, this is often unrealistic due to intractable transmission costs and security concerns. In this paper, we propose a method named ASCII for an agent to improve its classification performance through assistance from other agents. The main idea is to iteratively interchange an ignorance value between 0 and 1 for each collated sample among agents, where the value represents the urgency of further assistance needed. The method is naturally suitable for privacy-aware, transmission-economical, and decentralized learning scenarios. The method is also general as it allows the agents to use arbitrary classifiers such as logistic regression, ensemble tree, and neural network, and they may be heterogeneous among agents. We demonstrate the proposed method with extensive experimental studies.
I/Q modulation classification is a unique pattern recognition problem as the data for each class varies in quality, quantified by signal to noise ratio (SNR), and has structure in the complex-plane. Previous work shows treating these samples as complex-valued signals and computing complex-valued convolutions within deep learning frameworks significantly increases the performance over comparable shallow CNN architectures. In this work, we claim state of the art performance by enabling high-capacity architectures containing residual and/or dense connections to compute complex-valued convolutions, with peak classification accuracy of 92.4% on a benchmark classification problem, the RadioML 2016.10a dataset. We show statistically significant improvements in all networks with complex convolutions for I/Q modulation classification. Complexity and inference speed analyses show models with complex convolutions substantially outperform architectures with a comparable number of parameters and comparable speed by over 10% in each case.
Automatic detection and classification of pavement distresses is critical in timely maintaining and rehabilitating pavement surfaces. With the evolution of deep learning and high performance computing, the feasibility of vision-based pavement defect assessments has significantly improved. In this study, the authors deploy state-of-the-art deep learning algorithms based on different network backbones to detect and characterize pavement distresses. The influence of different backbone models such as CSPDarknet53, Hourglass-104 and EfficientNet were studied to evaluate their classification performance. The models were trained using 21,041 images captured across urban and rural streets of Japan, Czech Republic and India. Finally, the models were assessed based on their ability to predict and classify distresses, and tested using F1 score obtained from the statistical precision and recall values. The best performing model achieved an F1 score of 0.58 and 0.57 on two test datasets released by the IEEE Global Road Damage Detection Challenge. The source code including the trained models are made available at [1].
The traditional binary classification framework constructs classifiers which may have good accuracy, but whose false positive and false negative error rates are not under users' control. In many cases, one of the errors is more severe and only the classifiers with the corresponding rate lower than the predefined threshold are acceptable. In this study, we combine binary classification with statistical hypothesis testing to control the target error rate of already trained classifiers. In particular, we show how to turn binary classifiers into statistical tests, calculate the classification p-values, and use them to limit the target error rate.
We present a method for classifying syntactic errors in learner language, namely errors whose correction alters the morphosyntactic structure of a sentence. The methodology builds on the established Universal Dependencies syntactic representation scheme, and provides complementary information to other error-classification systems. Unlike existing error classification methods, our method is applicable across languages, which we showcase by producing a detailed picture of syntactic errors in learner English and learner Russian. We further demonstrate the utility of the methodology for analyzing the outputs of leading Grammatical Error Correction (GEC) systems.
There has been empirical evidence of presence of non-linearity and chaos at the level of single neurons in biological neural networks. The properties of chaotic neurons inspires us to employ them in artificial learning systems. Here, we propose a Neurochaos Learning (NL) architecture, where the neurons used to extract features from data are 1D chaotic maps. ChaosFEX+SVM, an instance of this NL architecture, is proposed as a hybrid combination of chaos and classical machine learning algorithm. We formally prove that a single layer of NL with a finite number of 1D chaotic neurons satisfies the Universal Approximation Theorem with an exact value for the number of chaotic neurons needed to approximate a discrete real valued function with finite support. This is made possible due to the topological transitivity property of chaos and the existence of uncountably infinite number of dense orbits for the chosen 1D chaotic map. The chaotic neurons in NL get activated under the presence of an input stimulus (data) and output a chaotic firing trajectory. From such chaotic firing trajectories of individual neurons of NL, we extract Firing Time, Firing Rate, Energy and Entropy that constitute ChaosFEX features. These ChaosFEX features are then fed to a Support Vector Machine with linear kernel for classification. The effectiveness of chaotic feature engineering performed by NL (ChaosFEX+SVM) is demonstrated for synthetic and real world datasets in the low and high training sample regimes. Specifically, we consider the problem of classification of genome sequences of SARS-CoV-2 from other coronaviruses (SARS-CoV-1, MERS-CoV and others). With just one training sample per class for 1000 random trials of training, we report an average macro F1-score > 0.99 for the classification of SARS-CoV-2 from SARS-CoV-1 genome sequences. Robustness of ChaosFEX features to additive noise is also demonstrated.
We consider the graph link prediction task, which is a classic graph analytical problem with many real-world applications. With the advances of deep learning, current link prediction methods commonly compute features from subgraphs centered at two neighboring nodes and use the features to predict the label of the link between these two nodes. In this formalism, a link prediction problem is converted to a graph classification task. In order to extract fixed-size features for classification, graph pooling layers are necessary in the deep learning model, thereby incurring information loss. To overcome this key limitation, we propose to seek a radically different and novel path by making use of the line graphs in graph theory. In particular, each node in a line graph corresponds to a unique edge in the original graph. Therefore, link prediction problems in the original graph can be equivalently solved as a node classification problem in its corresponding line graph, instead of a graph classification task. Experimental results on fourteen datasets from different applications demonstrate that our proposed method consistently outperforms the state-of-the-art methods, while it has fewer parameters and high training efficiency.
Sufficient dimension reduction is used pervasively as a supervised dimension reduction approach. Most existing sufficient dimension reduction methods are developed for data with a continuous response and may have an unsatisfactory performance for the categorical response, especially for the binary-response. To address this issue, we propose a novel estimation method of sufficient dimension reduction subspace (SDR subspace) using optimal transport. The proposed method, named principal optimal transport direction (POTD), estimates the basis of the SDR subspace using the principal directions of the optimal transport coupling between the data respecting different response categories. The proposed method also reveals the relationship among three seemingly irrelevant topics, i.e., sufficient dimension reduction, support vector machine, and optimal transport. We study the asymptotic properties of POTD and show that in the cases when the class labels contain no error, POTD estimates the SDR subspace exclusively. Empirical studies show POTD outperforms most of the state-of-the-art linear dimension reduction methods.
Malware detection and classification is a challenging problem and an active area of research. Traditional machine learning methods depend almost entirely on the ability to extract a set of discriminative features into which characterize malware. However, this feature engineering process is very time consuming. On the contrary, deep learning methods replace manual feature engineering by a system that performs both feature extraction and classification from raw data at once. Despite that, a major shortfall of these methods is their inhability to consider multiple disparate sources of information when performing classification, leading them to perform poorly when compared to multimodal approaches. In this work, we introduce Orthrus, a new bimodal approach to categorize malware into families based on deep learning. Orthrus combines two modalities of data: (1) the byte sequence representing the malware’s binary content, and (2) the assembly language instructions extracted from the assembly language source code of malware, and performs automatic feature learning and classification with a convolutional neural network. The idea is to benefit from multiple feature types to reflect malware’s characteristics. The experiments carried on the Microsoft Malware Classification Challenge dataset show that our proposed solution achieves higher classification performance than deep learning approaches in the literature and n-gram based methods.
Malware detection and classification is a challenging problem and an active area of research. Particular challenges include how to best treat and preprocess malicious executables in order to feed machine learning algorithms. Novel approaches in the literature treat an executable as a sequence of bytes or as a sequence of assembly language instructions. However, in those approaches the hierarchical structure of programs is not taken into consideration. An executable exhibits various levels of spatial correlation. Adjacent code instructions are correlated spatially but that is not necessarily the case. Function calls and jump commands transfer the control of the program to a different point in the instruction stream. Furthermore, these discontinuities are maintained when treating the binary as a sequence of byte values. In addition, functions might be arranged randomly if addresses are correctly reorganized. To address these issues we propose a Hierarchical Convolutional Network (HCN) for malware classification. It has two levels of convolutional blocks applied at the mnemonic-level and at the function-level, enabling us to extract n-gram like features from both levels when constructing the malware representation. We validate our HCN method on the dataset released for the Microsoft Malware Classification Challenge, outperforming almost every deep learning method in the literature.
Graph convolutional networks learn effective node embeddings that have proven to be useful in achieving high-accuracy prediction results in semi-supervised learning tasks, such as node classification. However, these networks suffer from the issue of over-smoothing and shrinking effect of the graph due in large part to the fact that they diffuse features across the edges of the graph using a linear Laplacian flow. This limitation is especially problematic for the task of node classification, where the goal is to predict the label associated with a graph node. To address this issue, we propose an anisotropic graph convolutional network for semi-supervised node classification by introducing a nonlinear function that captures informative features from nodes, while preventing oversmoothing. The proposed framework is largely motivated by the good performance of anisotropic diffusion in image and geometry processing, and learns nonlinear representations based on local graph structure and node features. The effectiveness of our approach is demonstrated on three citation networks and two image datasets, achieving better or comparable classification accuracy results compared to the standard baseline methods.
Robust learning from noisy demonstrations is a practical but highly challenging problem in imitation learning. In this paper, we first theoretically show that robust imitation learning can be achieved by optimizing a classification risk with a symmetric loss. Based on this theoretical finding, we then propose a new imitation learning method that optimizes the classification risk by effectively combining pseudo-labeling with co-training. Unlike existing methods, our method does not require additional labels or strict assumptions about noise distributions. Experimental results on continuous-control benchmarks show that our method is more robust compared to state-of-the-art methods.
Hierarchical multi-label classification (HMC) is a challenging classification task extending standard multi-label classification problems by imposing a hierarchy constraint on the classes. In this paper, we propose C-HMCNN(h), a novel approach for HMC problems, which, given a network h for the underlying multi-label classification problem, exploits the hierarchy information in order to produce predictions coherent with the constraint and improve performance. We conduct an extensive experimental analysis showing the superior performance of C-HMCNN(h) when compared to state-of-the-art models.
Automatic visual fire detection is used to complement traditional fire detection sensor systems (smoke/heat). In this work, we investigate different Convolutional Neural Network (CNN) architectures and their variants for the non-temporal real-time bounds detection of fire pixel regions in video (or still) imagery. Two reduced complexity compact CNN architectures (NasNet-A-OnFire and ShuffleNetV2-OnFire) are proposed through experimental analysis to optimise the computational efficiency for this task. The results improve upon the current state-of-the-art solution for fire detection, achieving an accuracy of 95% for full-frame binary classification and 97% for superpixel localisation. We notably achieve a classification speed up by a factor of 2.3x for binary classification and 1.3x for superpixel localisation, with runtime of 40 fps and 18 fps respectively, outperforming prior work in the field presenting an efficient, robust and real-time solution for fire region detection. Subsequent implementation on low-powered devices (Nvidia Xavier-NX, achieving 49 fps for full-frame classification via ShuffleNetV2-OnFire) demonstrates our architectures are suitable for various real-world deployment applications.
This paper presents an novel object type classification method for automotive applications which uses deep learning with radar reflections. The method provides object class information such as pedestrian, cyclist, car, or non-obstacle. The method is both powerful and efficient, by using a light-weight deep learning approach on reflection level radar data. It fills the gap between low-performant methods of handcrafted features and high-performant methods with convolutional neural networks. The proposed network exploits the specific characteristics of radar reflection data: It handles unordered lists of arbitrary length as input and it combines both extraction of local and global features. In experiments with real data the proposed network outperforms existing methods of handcrafted or learned features. An ablation study analyzes the impact of the proposed global context layer.