This paper proposes a new strategy for learning powerful cross-modal embeddings for audio-to-video synchronization. Here, we set up the problem as one of cross-modal retrieval, where the objective is to find the most relevant audio segment given a short video clip. The method builds on the recent advances in learning representations from cross-modal self-supervision. The main contributions of this paper are as follows: (1) we propose a new learning strategy where the embeddings are learnt via a multi-way matching problem, as opposed to a binary classification (matching or non-matching) problem as proposed by recent papers; (2) we demonstrate that performance of this method far exceeds the existing baselines on the synchronization task; (3) we use the learnt embeddings for visual speech recognition in self-supervision, and show that the performance matches the representations learnt end-to-end in a fully-supervised manner.
Bottlenecks of binary classification from positive and unlabeled data (PU classification) are the requirements that given unlabeled patterns are drawn from the test marginal distribution, and the penalty of the false positive error is identical to the false negative error. However, such requirements are often not fulfilled in practice. In this paper, we generalize PU classification to the class prior shift and asymmetric error scenarios. Based on the analysis of the Bayes optimal classifier, we show that given a test class prior, PU classification under class prior shift is equivalent to PU classification with asymmetric error. Then, we propose two different frameworks to handle these problems, namely, a risk minimization framework and density ratio estimation framework. Finally, we demonstrate the effectiveness of the proposed frameworks and compare both frameworks through experiments using benchmark datasets.
Prenatal diagnosis of tetralogy of Fallot (TOF) and hypoplastic left heart syndrome (HLHS), two serious congenital heart defects, improves outcomes and can in some cases facilitate in utero interventions. In practice, however, the fetal diagnosis rate for these lesions is only 30-50 percent in community settings. Improving fetal diagnosis of congenital heart disease is therefore critical. Deep learning is a cutting-edge machine learning technique for finding patterns in images but has not yet been applied to prenatal diagnosis of congenital heart disease. Using 685 retrospectively collected echocardiograms from fetuses 18-24 weeks of gestational age from 2000-2018, we trained convolutional and fully-convolutional deep learning models in a supervised manner to (i) identify the five canonical screening views of the fetal heart and (ii) segment cardiac structures to calculate fetal cardiac biometrics. We then trained models to distinguish by view between normal hearts, TOF, and HLHS. In a holdout test set of images, F-score for identification of the five most important fetal cardiac views was 0.95. Binary classification of unannotated cardiac views of normal heart vs. TOF reached an overall sensitivity of 75% and a specificity of 76%, while normal vs. HLHS reached a sensitivity of 100% and specificity of 90%, both well above average diagnostic rates for these lesions. Furthermore, segmentation-based measurements for cardiothoracic ratio (CTR), cardiac axis (CA), and ventricular fractional area change (FAC) were compatible with clinically measured metrics for normal, TOF, and HLHS hearts. Thus, using guideline-recommended imaging, deep learning models can significantly improve detection of fetal congenital heart disease compared to the common standard of care.
While a lot of progress has been made in recent years, the dynamics of learning in deep nonlinear neural networks remain to this day largely misunderstood. In this work, we study the case of binary classification and prove various properties of learning in such networks under strong assumptions such as linear separability of the data. Extending existing results from the linear case, we confirm empirical observations by proving that the classification error also follows a sigmoidal shape in nonlinear architectures. We show that given proper initialization, learning expounds parallel independent modes and that certain regions of parameter space might lead to failed training. We also demonstrate that input norm and features' frequency in the dataset lead to distinct convergence speeds which might shed some light on the generalization capabilities of deep neural networks. We provide a comparison between the dynamics of learning with cross-entropy and hinge losses, which could prove useful to understand recent progress in the training of generative adversarial networks. Finally, we identify a phenomenon that we baptize gradient starvation where the most frequent features in a dataset prevent the learning of other less frequent but equally informative features.
Many of the best statistical classification algorithms are binary classifiers that can only distinguish between one of two classes. The number of possible ways of generalizing binary classification to multi-class increases exponentially with the number of classes. There is some indication that the best method will depend on the dataset. Hence, we are particularly interested in data-driven solution design, whether based on prior considerations or on empirical examination of the data. Here we demonstrate how a recursive control language can be used to describe a multitude of different partitioning strategies in multi-class classification, including those in most common use. We use it both to manually construct new partitioning configurations as well as to examine those that have been automatically designed. Eight different strategies were tested on eight different datasets using a support vector machine (SVM) as the base binary classifier. Numerical results suggest that a one-size-fits-all solution consisting of one-versus-one is appropriate for most datasets. Three datasets showed better accuracy using different methods. The best solution for the most improved dataset exploited a property of the data to produce an uncertainty coefficient 36\% higher (0.016 absolute gain) than one-vs.-one. For the same dataset, an adaptive solution that empirically examined the data was also more accurate than one-vs.-one while being faster.
Nested dichotomies are used as a method of transforming a multiclass classification problem into a series of binary problems. A tree structure is induced that recursively splits the set of classes into subsets, and a binary classification model learns to discriminate between the two subsets of classes at each node. In this paper, we demonstrate that these nested dichotomies typically exhibit poor probability calibration, even when the base binary models are well calibrated. We also show that this problem is exacerbated when the binary models are poorly calibrated. We discuss the effectiveness of different calibration strategies and show that accuracy and log-loss can be significantly improved by calibrating both the internal base models and the full nested dichotomy structure, especially when the number of classes is high.
We study the sample complexity of semi-supervised learning (SSL) and introduce new assumptions based on the mismatch between a mixture model learned from unlabeled data and the true mixture model induced by the (unknown) class conditional distributions. Under these assumptions, we establish an $\Omega(K\log K)$ labeled sample complexity bound without imposing parametric assumptions, where $K$ is the number of classes. Our results suggest that even in nonparametric settings it is possible to learn a near-optimal classifier using only a few labeled samples. Unlike previous theoretical work which focuses on binary classification, we consider general multiclass classification ($K>2$), which requires solving a difficult permutation learning problem. This permutation defines a classifier whose classification error is controlled by the Wasserstein distance between mixing measures, and we provide finite-sample results characterizing the behaviour of the excess risk of this classifier. Finally, we describe three algorithms for computing these estimators based on a connection to bipartite graph matching, and perform experiments to illustrate the superiority of the MLE over the majority vote estimator.
The Wireless Capsule Endoscopy (WCE) is becoming a popular way of screening gastrointestinal system diseases and cancer. However, the time-consuming process in inspecting WCE data limits its applications and increases the cost of examinations. This paper considers WCE-based gastric ulcer detection, in which the major challenge is to detect the lesions in a local region. We propose an approach named infinite curriculum learning, which generalizes curriculum learning to an infinite sampling space by approximately measuring the difficulty of each patch by its scale. This allows us to adapt our model from local patches to global images gradually, leading to a consistent accuracy gain. Experiments are performed on a large dataset with more than 3 million WCE images. Our approach achieves a binary classification accuracy of 87%, and is able to detect some lesions mis-annotated by the physicians. In a real-world application, our approach can reduce the workload of a physician by 90%-98% in gastric ulcer screening.
We propose Information-Theoretic Active Learning (ITAL), a novel batch-mode active learning method for binary classification, and apply it for acquiring meaningful user feedback in the context of content-based image retrieval. Instead of combining different heuristics such as uncertainty, diversity, or density, our method is based on maximizing the mutual information between the predicted relevance of the images and the expected user feedback regarding the selected batch. We propose suitable approximations to this computationally demanding problem and also integrate an explicit model of user behavior that accounts for possible incorrect labels and unnameable instances. Furthermore, our approach does not only take the structure of the data but also the expected model output change caused by the user feedback into account. In contrast to other methods, ITAL turns out to be highly flexible and provides state-of-the-art performance across various datasets, such as MIRFLICKR and ImageNet.
Existing work on automated hate speech detection typically focuses on binary classification or on differentiating among a small set of categories. In this paper, we propose a novel method on a fine-grained hate speech classification task, which focuses on differentiating among 40 hate groups of 13 different hate group categories. We first explore the Conditional Variational Autoencoder (CVAE) as a discriminative model and then extend it to a hierarchical architecture to utilize the additional hate category information for more accurate prediction. Experimentally, we show that incorporating the hate category information for training can significantly improve the classification performance and our proposed model outperforms commonly-used discriminative models.
Consider a query-based data acquisition problem that aims to recover the values of $k$ binary variables from parity (XOR) measurements of chosen subsets of the variables. Assume the response model where only a randomly selected subset of the measurements is received. We propose a method for designing a sequence of queries so that the variables can be identified with high probability using as few ($n$) measurements as possible. We define the query difficulty $\bar{d}$ as the average size of the query subsets and the sample complexity $n$ as the minimum number of measurements required to attain a given recovery accuracy. We obtain fundamental trade-offs between recovery accuracy, query difficulty, and sample complexity. In particular, the necessary and sufficient sample complexity required for recovering all $k$ variables with high probability is $n = c_0 \max\{k, (k \log k)/\bar{d}\}$ and the sample complexity for recovering a fixed proportion $(1-\delta)k$ of the variables for $\delta=o(1)$ is $n = c_1\max\{k, (k \log(1/\delta))/\bar{d}\}$, where $c_0, c_1>0$.
We introduce the chi-square test neural network: a single hidden layer backpropagation neural network using chi-square test theorem to redefine the cost function and the error function. The weights and thresholds are modified using standard backpropagation algorithm. The proposed approach has the advantage of making consistent data distribution over training and testing sets. It can be used for binary classification. The experimental results on real world data sets indicate that the proposed algorithm can significantly improve the classification accuracy comparing to related approaches.
Many optimization techniques evaluate solutions consecutively, where the next candidate for evaluation is determined by the results of previous evaluations. For example, these include iterative methods, "black box" optimization algorithms, simulated annealing, evolutionary algorithms and tabu search, to name a few. When solving an optimization problem, these algorithms evaluate a large number of solutions, which raises the following question: Is it possible to learn something about the optimum using these solutions? In this paper, we define this "learning" question in terms of a logistic regression model and explore its predictive accuracy computationally. The proposed model uses a collection of solutions to predict the components of the optimal solutions. To illustrate the utility of such predictions, we embed the logistic regression model into the tabu search algorithm for job shop scheduling problem. The resulting framework is simple to implement, yet provides a significant boost to the performance of the standard tabu search.
This paper presents a novel approach to the segmentation of orthographic word forms in contemporary Hebrew, focusing purely on splitting without carrying out morphological analysis or disambiguation. Casting the analysis task as character-wise binary classification and using adjacent character and word-based lexicon-lookup features, this approach achieves over 98% accuracy on the benchmark SPMRL shared task data for Hebrew, and 97% accuracy on a new out of domain Wikipedia dataset, an improvement of ~4% and 5% over previous state of the art performance.
The electroencephalogram (EEG) is the most popular form of input for brain computer interfaces (BCIs). However, it can be easily contaminated by various artifacts and noise, e.g., eye blink, muscle activities, powerline noise, etc. Therefore, the EEG signals are often filtered both spatially and temporally to increase the signal-to-noise ratio before they are fed into a machine learning algorithm for recognition. This paper considers spatial filtering, particularly, the common spatial pattern (CSP) filters for EEG classification. In binary classification, CSP seeks a set of filters to maximize the variance for one class while minimizing it for the other. We first introduce the traditional solution, and then a new solution based on a slightly different objective function. We performed comprehensive experiments on motor imagery to compare the two approaches, and found that generally the traditional CSP solution still gives better results. We also showed that adding regularization to the covariance matrices can improve the final classification performance, no matter which objective function is used.
We consider the problem in precision health of grouping people into subpopulations based on their degree of vulnerability to a risk factor. These subpopulations cannot be discovered with traditional clustering techniques because their quality is evaluated with a supervised metric: the ease of modeling a response variable over observations within them. Instead, we apply the supervised cadre model (SCM), which does use this metric. We extend the SCM formalism so that it may be applied to multivariate regression and binary classification problems. We also develop a way to use conditional entropy to assess the confidence in the process by which a subject is assigned their cadre. Using the SCM, we generalize the environment-wide association study (EWAS) workflow to be able to model heterogeneity in population risk. In our EWAS, we consider more than two hundred environmental exposure factors and find their association with diastolic blood pressure, systolic blood pressure, and hypertension. This requires adapting the SCM to be applicable to data generated by a complex survey design. After correcting for false positives, we found 25 exposure variables that had a significant association with at least one of our response variables. Eight of these were significant for a discovered subpopulation but not for the overall population. Some of these associations have been identified by previous researchers, while others appear to be novel. We examine several discovered subpopulations in detail, and we find that they are interpretable and that they suggest further research questions.
Neural networks with REctified Linear Unit (ReLU) activation functions (a.k.a. ReLU networks) have achieved great empirical success in various domains. Nonetheless, existing results for learning ReLU networks either pose assumptions on the underlying data distribution being e.g. Gaussian, or require the network size and/or training size to be sufficiently large. In this context, the problem of learning a two-layer ReLU network is approached in a binary classification setting, where the data are linearly separable and a hinge loss criterion is adopted. Leveraging the power of random noise perturbation, this paper presents a novel stochastic gradient descent (SGD) algorithm, which can \emph{provably} train any single-hidden-layer ReLU network to attain global optimality, despite the presence of infinitely many bad local minima, maxima, and saddle points in general. This result is the first of its kind, requiring no assumptions on the data distribution, training/network size, or initialization. Convergence of the resultant iterative algorithm to a global minimum is analyzed by establishing both an upper bound and a lower bound on the number of non-zero updates to be performed. Moreover, generalization guarantees are developed for ReLU networks trained with the novel SGD leveraging classic compression bounds. These guarantees highlight a key difference (at least in the worst case) between reliably learning a ReLU network as well as a leaky ReLU network in terms of sample complexity. Numerical tests using both synthetic data and real images validate the effectiveness of the algorithm and the practical merits of the theory.
Sentence encoders are typically trained on language modeling tasks with large unlabeled datasets. While these encoders achieve state-of-the-art results on many sentence-level tasks, they are difficult to train with long training cycles. We introduce fake sentence detection as a new training task for learning sentence encoders. We automatically generate fake sentences by corrupting original sentences from a source collection and train the encoders to produce representations that are effective at detecting fake sentences. This binary classification task turns to be quite efficient for training sentence encoders. We compare a basic BiLSTM encoder trained on this task with a strong sentence encoding models (Skipthought and FastSent) trained on a language modeling task. We find that the BiLSTM trains much faster on fake sentence detection (20 hours instead of weeks) using smaller amounts of data (1M instead of 64M sentences). Further analysis shows the learned representations capture many syntactic and semantic properties expected from good sentence representations.
We present a new recurrent neural network topology to enhance state-of-the-art machine learning systems by incorporating a broader context. Our approach overcomes recent limitations with extended narratives through a multi-layered computational approach to generate an abstract context representation. Therefore, the developed system captures the narrative on word-level, sentence-level, and context-level. Through the hierarchical set-up, our proposed model summarizes the most salient information on each level and creates an abstract representation of the extended context. We subsequently use this representation to enhance neural language processing systems on the task of semantic error detection. To show the potential of the newly introduced topology, we compare the approach against a context-agnostic set-up including a standard neural language model and a supervised binary classification network. The performance measures on the error detection task show the advantage of the hierarchical context-aware topologies, improving the baseline by 12.75% relative for unsupervised models and 20.37% relative for supervised models.
Supervised learning based methods for source localization, being data driven, can be adapted to different acoustic conditions via training and have been shown to be robust to adverse acoustic environments. In this paper, a convolutional neural network (CNN) based supervised learning method for estimating the direction-of-arrival (DOA) of multiple speakers is proposed. Multi-speaker DOA estimation is formulated as a multi-class multi-label classification problem, where the assignment of each DOA label to the input feature is treated as a separate binary classification problem. The phase component of the short-time Fourier transform (STFT) coefficients of the received microphone signals are directly fed into the CNN, and the features for DOA estimation are learnt during training. Utilizing the assumption of disjoint speaker activity in the STFT domain, a novel method is proposed to train the CNN with synthesized noise signals. Through experimental evaluation with both simulated and measured acoustic impulse responses, the ability of the proposed DOA estimation approach to adapt to unseen acoustic conditions and its robustness to unseen noise type is demonstrated. Through additional empirical investigation, it is also shown that with an array of M microphones our proposed framework yields the best localization performance with M-1 convolution layers. The ability of the proposed method to accurately localize speakers in a dynamic acoustic scenario with varying number of sources is also shown.