It is common practice of the outlier mining community to repurpose classification datasets toward evaluating various detection models. To that end, often a binary classification dataset is used, where samples from one of the classes is designated as the inlier samples, and the other class is substantially down-sampled to create the ground-truth outlier samples. Graph-level outlier detection (GLOD) is rarely studied but has many potentially influential real-world applications. In this study, we identify an intriguing issue with repurposing graph classification datasets for GLOD. We find that ROC-AUC performance of the models changes significantly (flips from high to very low, even worse than random) depending on which class is down-sampled. Interestingly, ROC-AUCs on these two variants approximately sum to 1 and their performance gap is amplified with increasing propagations for a certain family of propagation based outlier detection models. We carefully study the graph embedding space produced by propagation based models and find two driving factors: (1) disparity between within-class densities which is amplified by propagation, and (2)overlapping support (mixing of embeddings) across classes. We also study other graph embedding methods and downstream outlier detectors, and find that the intriguing performance flip issue still widely exists but which version of the downsample achieves higher performance may vary. Thoughtful analysis over comprehensive results further deeper our understanding of the established issue.
Skew-Gaussian processes (SkewGPs) extend the multivariate Unified Skew-Normal distributions over finite dimensional vectors to distribution over functions. SkewGPs are more general and flexible than Gaussian processes, as SkewGPs may also represent asymmetric distributions. In a recent contribution we showed that SkewGP and probit likelihood are conjugate, which allows us to compute the exact posterior for non-parametric binary classification and preference learning. In this paper, we generalize previous results and we prove that SkewGP is conjugate with both the normal and affine probit likelihood, and more in general, with their product. This allows us to (i) handle classification, preference, numeric and ordinal regression, and mixed problems in a unified framework; (ii) derive closed-form expression for the corresponding posterior distributions. We show empirically that the proposed framework based on SkewGP provides better performance than Gaussian processes in active learning and Bayesian (constrained) optimization. These two tasks are fundamental for design of experiments and in Data Science.
Unsupervised anomalous sound detection is concerned with identifying sounds that deviate from what is defined as 'normal', without explicitly specifying the types of anomalies. A significant obstacle is the diversity and rareness of outliers, which typically prevent us from collecting a representative set of anomalous sounds. As a consequence, most anomaly detection methods use unsupervised rather than supervised machine learning methods. Nevertheless, we will show that anomalous sound detection can be effectively framed as a supervised classification problem if the set of anomalous samples is carefully substituted with what we call proxy outliers. Candidates for proxy outliers are available in abundance as they potentially include all recordings that are neither normal nor abnormal sounds. We experiment with the machine condition monitoring data set of the 2020's DCASE Challenge and find proxy outliers with matching recording conditions and high similarity to the target sounds particularly informative. If no data with similar sounds and matching recording conditions is available, data sets with a larger diversity in these two dimensions are preferable. Our models based on supervised training with proxy outliers achieved rank three in Task 2 of the DCASE2020 Challenge.
Weakly supervised disease classification of CT imaging suffers from poor localization owing to case-level annotations, where even a positive scan can hold hundreds to thousands of negative slices along multiple planes. Furthermore, although deep learning segmentation and classification models extract distinctly unique combinations of anatomical features from the same target class(es), they are typically seen as two independent processes in a computer-aided diagnosis (CAD) pipeline, with little to no feature reuse. In this research, we propose a medical classifier that leverages the semantic structural concepts learned via multi-resolution segmentation feature maps, to guide weakly supervised 3D classification of chest CT volumes. Additionally, a comparative analysis is drawn across two different types of feature aggregation to explore the vast possibilities surrounding feature fusion. Using a dataset of 1593 scans labeled on a case-level basis via rule-based model, we train a dual-stage convolutional neural network (CNN) to perform organ segmentation and binary classification of four representative diseases (emphysema, pneumonia/atelectasis, mass and nodules) in lungs. The baseline model, with separate stages for segmentation and classification, results in AUC of 0.791. Using identical hyperparameters, the connected architecture using static and dynamic feature aggregation improves performance to AUC of 0.832 and 0.851, respectively. This study advances the field in two key ways. First, case-level report data is used to weakly supervise a 3D CT classifier of multiple, simultaneous diseases for an organ. Second, segmentation and classification models are connected with two different feature aggregation strategies to enhance the classification performance.
One reason pretraining on self-supervised linguistic tasks is effective is that it teaches models features that are helpful for language understanding. However, we want pretrained models to learn not only to represent linguistic features, but also to use those features preferentially during fine-turning. With this goal in mind, we introduce a new English-language diagnostic set called MSGS (the Mixed Signals Generalization Set), which consists of 20 ambiguous binary classification tasks that we use to test whether a pretrained model prefers linguistic or surface generalizations during fine-tuning. We pretrain RoBERTa models from scratch on quantities of data ranging from 1M to 1B words and compare their performance on MSGS to the publicly available RoBERTa-base. We find that models can learn to represent linguistic features with little pretraining data, but require far more data to learn to prefer linguistic generalizations over surface ones. Eventually, with about 30B words of pretraining data, RoBERTa-base does demonstrate a linguistic bias with some regularity. We conclude that while self-supervised pretraining is an effective way to learn helpful inductive biases, there is likely room to improve the rate at which models learn which features matter.
Systematic reviews, which entail the extraction of data from large numbers of scientific documents, are an ideal avenue for the application of machine learning. They are vital to many fields of science and philanthropy, but are very time-consuming and require experts. Yet the three main stages of a systematic review are easily done automatically: searching for documents can be done via APIs and scrapers, selection of relevant documents can be done via binary classification, and extraction of data can be done via sequence-labelling classification. Despite the promise of automation for this field, little research exists that examines the various ways to automate each of these tasks. We construct a pipeline that automates each of these aspects, and experiment with many human-time vs. system quality trade-offs. We test the ability of classifiers to work well on small amounts of data and to generalise to data from countries not represented in the training data. We test different types of data extraction with varying difficulty in annotation, and five different neural architectures to do the extraction. We find that we can get surprising accuracy and generalisability of the whole pipeline system with only 2 weeks of human-expert annotation, which is only 15% of the time it takes to do the whole review manually and can be repeated and extended to new data with no additional effort.
In forming learning objectives, one oftentimes needs to aggregate a set of individual values to a single output. Such cases occur in the aggregate loss, which combines individual losses of a learning model over each training sample, and in the individual loss for multi-label learning, which combines prediction scores over all class labels. In this work, we introduce the sum of ranked range (SoRR) as a general approach to form learning objectives. A ranked range is a consecutive sequence of sorted values of a set of real numbers. The minimization of SoRR is solved with the difference of convex algorithm (DCA). We explore two applications in machine learning of the minimization of the SoRR framework, namely the AoRR aggregate loss for binary classification and the TKML individual loss for multi-label/multi-class classification. Our empirical results highlight the effectiveness of the proposed optimization framework and demonstrate the applicability of proposed losses using synthetic and real datasets.
Table entailment, the binary classification task of finding if a sentence is supported or refuted by the content of a table, requires parsing language and table structure as well as numerical and discrete reasoning. While there is extensive work on textual entailment, table entailment is less well studied. We adapt TAPAS (Herzig et al., 2020), a table-based BERT model, to recognize entailment. Motivated by the benefits of data augmentation, we create a balanced dataset of millions of automatically created training examples which are learned in an intermediate step prior to fine-tuning. This new data is not only useful for table entailment, but also for SQA (Iyyer et al., 2017), a sequential table QA task. To be able to use long examples as input of BERT models, we evaluate table pruning techniques as a pre-processing step to drastically improve the training and prediction efficiency at a moderate drop in accuracy. The different methods set the new state-of-the-art on the TabFact (Chen et al., 2020) and SQA datasets.
Feature selection is an essential step in data science pipelines to reduce the complexity associated with large datasets. While much research on this topic focuses on optimizing predictive performance, few studies investigate stability in the context of the feature selection process. In this study, we present the Repeated Elastic Net Technique (RENT) for Feature Selection. RENT uses an ensemble of generalized linear models with elastic net regularization, each trained on distinct subsets of the training data. The feature selection is based on three criteria evaluating the weight distributions of features across all elementary models. This fact leads to the selection of features with high stability that improve the robustness of the final model. Furthermore, unlike established feature selectors, RENT provides valuable information for model interpretation concerning the identification of objects in the data that are difficult to predict during training. In our experiments, we benchmark RENT against six established feature selectors on eight multivariate datasets for binary classification and regression. In the experimental comparison, RENT shows a well-balanced trade-off between predictive performance and stability. Finally, we underline the additional interpretational value of RENT with an exploratory post-hoc analysis of a healthcare dataset.
In this paper, we study the task of selecting the optimal response given a user and system utterance history in retrieval-based multi-turn dialog systems. Recently, pre-trained language models (e.g., BERT, RoBERTa, and ELECTRA) showed significant improvements in various natural language processing tasks. This and similar response selection tasks can also be solved using such language models by formulating the tasks as dialog--response binary classification tasks. Although existing works using this approach successfully obtained state-of-the-art results, we observe that language models trained in this manner tend to make predictions based on the relatedness of history and candidates, ignoring the sequential nature of multi-turn dialog systems. This suggests that the response selection task alone is insufficient for learning temporal dependencies between utterances. To this end, we propose utterance manipulation strategies (UMS) to address this problem. Specifically, UMS consist of several strategies (i.e., insertion, deletion, and search), which aid the response selection model towards maintaining dialog coherence. Further, UMS are self-supervised methods that do not require additional annotation and thus can be easily incorporated into existing approaches. Extensive evaluation across multiple languages and models shows that UMS are highly effective in teaching dialog consistency, which leads to models pushing the state-of-the-art with significant margins on multiple public benchmark datasets.
We propose a new binary classification model called Phase Separation Binary Classifier (PSBC). It consists of a discretization of a nonlinear reaction-diffusion equation coupled with an Ordinary Differential Equation, and is inspired by fluids behavior, namely, on how binary fluids phase separate. Thus, parameters and hyperparameters have physical meaning, whose effects are studied in several different scenarios. PSBC's equations can be seen as a dynamical system whose coefficients are trainable weights, with a similar architecture to that of a Recurrent Neural Network. As such, forward propagation amounts to an initial value problem. Boundary conditions are also present, bearing similarity with figure padding techniques in Computer Vision. Model compression is exploited in several ways, with weight sharing taking place both across and within layers. The model is tested on pairs of digits of the classical MNIST database. An associated multiclass classifier is also constructed using a combination of Ensemble Learning and one versus one techniques. It is also shown how the PSBC can be combined with other methods - like aggregation and PCA - in order to construct better binary classifiers. The role of boundary conditions and viscosity is thoroughly studied in the case of digits ``0'' and ``1''.
Machine learning (ML) offers a collection of powerful approaches for detecting and modeling associations, often applied to data having a large number of features and/or complex associations. Currently, there are many tools to facilitate implementing custom ML analyses (e.g. scikit-learn). Interest is also increasing in automated ML packages, which can make it easier for non-experts to apply ML and have the potential to improve model performance. ML permeates most subfields of biomedical research with varying levels of rigor and correct usage. Tremendous opportunities offered by ML are frequently offset by the challenge of assembling comprehensive analysis pipelines, and the ease of ML misuse. In this work we have laid out and assembled a complete, rigorous ML analysis pipeline focused on binary classification (i.e. case/control prediction), and applied this pipeline to both simulated and real world data. At a high level, this 'automated' but customizable pipeline includes a) exploratory analysis, b) data cleaning and transformation, c) feature selection, d) model training with 9 established ML algorithms, each with hyperparameter optimization, and e) thorough evaluation, including appropriate metrics, statistical analyses, and novel visualizations. This pipeline organizes the many subtle complexities of ML pipeline assembly to illustrate best practices to avoid bias and ensure reproducibility. Additionally, this pipeline is the first to compare established ML algorithms to 'ExSTraCS', a rule-based ML algorithm with the unique capability of interpretably modeling heterogeneous patterns of association. While designed to be widely applicable we apply this pipeline to an epidemiological investigation of established and newly identified risk factors for pancreatic cancer to evaluate how different sources of bias might be handled by ML algorithms.
Malignant melanoma (MM) is one of the deadliest types of skin cancer. Analysing dermatoscopic images plays an important role in the early detection of MM and other pigmented skin lesions. Among different computer-based methods, deep learning-based approaches and in particular convolutional neural networks have shown excellent classification and segmentation performances for dermatoscopic skin lesion images. These models can be trained end-to-end without requiring any hand-crafted features. However, the effect of using lesion segmentation information on classification performance has remained an open question. In this study, we explicitly investigated the impact of using skin lesion segmentation masks on the performance of dermatoscopic image classification. To do this, first, we developed a baseline classifier as the reference model without using any segmentation masks. Then, we used either manually or automatically created segmentation masks in both training and test phases in different scenarios and investigated the classification performances. Evaluated on the ISIC 2017 challenge dataset which contained two binary classification tasks (i.e. MM vs. all and seborrheic keratosis (SK) vs. all) and based on the derived area under the receiver operating characteristic curve scores, we observed four main outcomes. Our results show that 1) using segmentation masks did not significantly improve the MM classification performance in any scenario, 2) in one of the scenarios (using segmentation masks for dilated cropping), SK classification performance was significantly improved, 3) removing all background information by the segmentation masks significantly degraded the overall classification performance, and 4) in case of using the appropriate scenario (using segmentation for dilated cropping), there is no significant difference of using manually or automatically created segmentation masks.
We describe our top-team solution to Task 1 for Hindi in the HASOC contest organised by FIRE 2019. The task is to identify hate speech and offensive language in Hindi. More specifically, it is a binary classification problem where a system is required to classify tweets into two classes: (a) \emph{Hate and Offensive (HOF)} and (b) \emph{Not Hate or Offensive (NOT)}. In contrast to the popular idea of pretraining word vectors (a.k.a. word embedding) with a large corpus from a general domain such as Wikipedia, we used a relatively small collection of relevant tweets (i.e. random and sarcasm tweets in Hindi and Hinglish) for pretraining. We trained a Convolutional Neural Network (CNN) on top of the pretrained word vectors. This approach allowed us to be ranked first for this task out of all teams. Our approach could easily be adapted to other applications where the goal is to predict class of a text when the provided context is limited.
Predicting a customer's propensity-to-pay at an early point in the revenue cycle can provide organisations many opportunities to improve the customer experience, reduce hardship and reduce the risk of impaired cash flow and occurrence of bad debt. With the advancements in data science; machine learning techniques can be used to build models to accurately predict a customer's propensity-to-pay. Creating effective machine learning models without access to large and detailed datasets presents some significant challenges. This paper presents a case-study, conducted on a dataset from an energy organisation, to explore the uncertainty around the creation of machine learning models that are able to predict residential customers entering financial hardship which then reduces their ability to pay energy bills. Incorrect predictions can result in inefficient resource allocation and vulnerable customers not being proactively identified. This study investigates machine learning models' ability to consider different contexts and estimate the uncertainty in the prediction. Seven models from four families of machine learning algorithms are investigated for their novel utilisation. A novel concept of utilising a Baysian Neural Network to the binary classification problem of propensity-to-pay energy bills is proposed and explored for deployment.
Motivated by the general robustness properties of the 01 loss we propose a single hidden layer 01 loss neural network trained with stochastic coordinate descent as a defense against adversarial attacks in machine learning. One measure of a model's robustness is the minimum distortion required to make the input adversarial. This can be approximated with the Boundary Attack (Brendel et. al. 2018) and HopSkipJump (Chen et. al. 2019) methods. We compare the minimum distortion of the 01 loss network to the binarized neural network and the standard sigmoid activation network with cross-entropy loss all trained with and without Gaussian noise on the CIFAR10 benchmark binary classification between classes 0 and 1. Both with and without noise training we find our 01 loss network to have the largest adversarial distortion of the three models by non-trivial margins. To further validate these results we subject all models to substitute model black box attacks under different distortion thresholds and find that the 01 loss network is the hardest to attack across all distortions. At a distortion of 0.125 both sigmoid activated cross-entropy loss and binarized networks have almost 0% accuracy on adversarial examples whereas the 01 loss network is at 40%. Even though both 01 loss and the binarized network use sign activations their training algorithms are different which in turn give different solutions for robustness. Finally we compare our network to simple convolutional models under substitute model black box attacks and find their accuracies to be comparable. Our work shows that the 01 loss network has the potential to defend against black box adversarial attacks better than convex loss and binarized networks.
We present a novel deep neural architecture for learning electroencephalogram (EEG). To learn the spatial information, our model first obtains the Riemannian mean and distance from spatial covariance matrices (SCMs) on a Riemannian manifold. We then project the spatial information onto a Euclidean space via tangent space learning. Following, two fully connected layers are used to learn the spatial information embeddings. Moreover, our proposed method learns the temporal information via differential entropy and logarithm power spectrum density features extracted from EEG signals in a Euclidean space using a deep long short-term memory network with a soft attention mechanism. To combine the spatial and temporal information, we use an effective fusion strategy, which learns attention weights applied to embedding-specific features for decision making. We evaluate our proposed framework on four public datasets across three popular EEG-related tasks, notably emotion recognition, vigilance estimation, and motor imagery classification, containing various types of tasks such as binary classification, multi-class classification, and regression. Our proposed architecture outperforms other methods on SEED-VIG, and approaches the state-of-the-art on the other three datasets (SEED, BCI-IV 2A, and BCI-IV 2B), showing the robustness of our framework in EEG representation learning. The source code of our paper is publicly available at https://github.com/guangyizhangbci/EEG_Riemannian.
The application of data mining, machine learning and artificial intelligence techniques in the field of diagnostics is not a new concept, and these techniques have been very successfully applied in a variety of applications, especially in dermatology and cancer research. But, in the case of medical problems that involve tests resulting in true or false (binary classification), the data generally has a class imbalance with samples majorly belonging to one class (ex: a patient undergoes a regular test and the results are false). Such disparity in data causes problems when trying to model predictive systems on the data. In critical applications like diagnostics, this class imbalance cannot be overlooked and must be given extra attention. In our research, we depict how we can handle this class imbalance through neural networks using a discriminative model and contrastive loss using a Siamese neural network structure. Such a model does not work on a probability-based approach to classify samples into labels. Instead it uses a distance-based approach to differentiate between samples classified under different labels. The code is available at https://tinyurl.com/DiscriminativeCHD/
Diabetic Retinopathy (DR) is one of the microvascular complications of Diabetes Mellitus, which remains as one of the leading causes of blindness worldwide. Computational models based on Convolutional Neural Networks represent the state of the art for the automatic detection of DR using eye fundus images. Most of the current work address this problem as a binary classification task. However, including the grade estimation and quantification of predictions uncertainty can potentially increase the robustness of the model. In this paper, a hybrid Deep Learning-Gaussian process method for DR diagnosis and uncertainty quantification is presented. This method combines the representational power of deep learning, with the ability to generalize from small datasets of Gaussian process models. The results show that uncertainty quantification in the predictions improves the interpretability of the method as a diagnostic support tool. The source code to replicate the experiments is publicly available at https://github.com/stoledoc/DLGP-DR-Diagnosis.
A common approach to medical image analysis on volumetric data uses deep 2D convolutional neural networks (CNNs). This is largely attributed to the challenges imposed by the nature of the 3D data: variable volume size, GPU exhaustion during optimization. However, dealing with the individual slices independently in 2D CNNs deliberately discards the depth information which results in poor performance for the intended task. Therefore, it is important to develop methods that not only overcome the heavy memory and computation requirements but also leverage the 3D information. To this end, we evaluate a set of volume uniformizing methods to address the aforementioned issues. The first method involves sampling information evenly from a subset of the volume. Another method exploits the full geometry of the 3D volume by interpolating over the z-axis. We demonstrate performance improvements using controlled ablation studies as well as put this approach to the test on the ImageCLEF Tuberculosis Severity Assessment 2019 benchmark. We report 73% area under curve (AUC) and binary classification accuracy (ACC) of 67.5% on the test set beating all methods which leveraged only image information (without using clinical meta-data) achieving 5-th position overall. All codes and models are made available at https://github.com/hasibzunair/uniformizing-3D.