In this paper, we propose FCHD-Fully Convolutional Head Detector, an end-to-end trainable head detection model. Our proposed architecture is a single fully convolutional network which is responsible for both bounding box prediction and classification. This makes our model lightweight with low inference time and memory requirements. Along with run-time, our model has better overall average precision (AP) which is achieved by selection of anchor sizes based on the effective receptive field of the network. This can be concluded from our experiments on several head detection datasets with varying head counts. We achieve an AP of 0.70 on a challenging head detection dataset which is comparable to some standard benchmarks. Along with this our model runs at 5 FPS on Nvidia Quadro M1000M for VGA resolution images. Code is available at https://github.com/aditya-vora/FCHD-Fully-Convolutional-Head-Detector.
We propose a novel linear discriminant analysis approach for the classification of high-dimensional matrix-valued data that commonly arises from imaging studies. Motivated by the equivalence of the conventional linear discriminant analysis and the ordinary least squares, we consider an efficient nuclear norm penalized regression that encourages a low-rank structure. Theoretical properties including a non-asymptotic risk bound and a rank consistency result are established. Simulation studies and an application to electroencephalography data show the superior performance of the proposed method over the existing approaches.
In Intelligent Tutoring System (ITS), tracing the student's knowledge state during learning has been studied for several decades in order to provide more supportive learning instructions. In this paper, we propose a novel model for knowledge tracing that i) captures students' learning ability and dynamically assigns students into distinct groups with similar ability at regular time intervals, and ii) combines this information with a Recurrent Neural Network architecture known as Deep Knowledge Tracing. Experimental results confirm that the proposed model is significantly better at predicting student performance than well known state-of-the-art techniques for student modelling.
In classic fair division problems such as cake cutting and rent division, envy-freeness requires that each individual (weakly) prefer his allocation to anyone else's. On a conceptual level, we argue that envy-freeness also provides a compelling notion of fairness for classification tasks. Our technical focus is the generalizability of envy-free classification, i.e., understanding whether a classifier that is envy free on a sample would be almost envy free with respect to the underlying distribution with high probability. Our main result establishes that a small sample is sufficient to achieve such guarantees, when the classifier in question is a mixture of deterministic classifiers that belong to a family of low Natarajan dimension.
In multilingual societies like the Indian subcontinent, use of code-switched languages is much popular and convenient for the users. In this paper, we study offense and abuse detection in the code-switched pair of Hindi and English (i.e. Hinglish), the pair that is the most spoken. The task is made difficult due to non-fixed grammar, vocabulary, semantics and spellings of Hinglish language. We apply transfer learning and make a LSTM based model for hate speech classification. This model surpasses the performance shown by the current best models to establish itself as the state-of-the-art in the unexplored domain of Hinglish offensive text classification.We also release our model and the embeddings trained for research purposes
In open set learning, a model must be able to generalize to novel classes when it encounters a sample that does not belong to any of the classes it has seen before. Open set learning poses a realistic learning scenario that is receiving growing attention. Existing studies on open set learning mainly focused on detecting novel classes, but few studies tried to model them for differentiating novel classes. In this paper, we recognize that novel classes should be different from each other, and propose distribution networks for open set learning that can model different novel classes based on probability distributions. We hypothesize that, through a certain mapping, samples from different classes with the same classification criterion should follow different probability distributions from the same distribution family. A deep neural network is learned to map the samples in the original feature space to a latent space where the distributions of known classes can be jointly learned with the network. We additionally propose a distribution parameter transfer and updating strategy for novel class modeling when a novel class is detected in the latent space. By novel class modeling, the detected novel classes can serve as known classes to the subsequent classification. Our experimental results on image datasets MNIST and CIFAR10 show that the distribution networks can detect novel classes accurately, and model them well for the subsequent classification tasks.
We study the problem of learning classifiers robust to universal adversarial perturbations. While prior work approaches this problem via robust optimization, adversarial training, or input transformation, we instead phrase it as a two-player zero-sum game. In this new formulation, both players simultaneously play the same game, where one player chooses a classifier that minimizes a classification loss whilst the other player creates an adversarial perturbation that increases the same loss when applied to every sample in the training set. By observing that performing a classification (respectively creating adversarial samples) is the best response to the other player, we propose a novel extension of a game-theoretic algorithm, namely fictitious play, to the domain of training robust classifiers. Finally, we empirically show the robustness and versatility of our approach in two defence scenarios where universal attacks are performed on several image classification datasets -- CIFAR10, CIFAR100 and ImageNet.
Machine learning methods such as convolutional neural networks (CNNs) are becoming an integral part of scientific research in many disciplines, spatial vector data often fail to be analyzed using these powerful learning methods because of its irregularities. With the aid of graph Fourier transform and convolution theorem, it is possible to convert the convolution as a point-wise product in Fourier domain and construct a learning architecture of CNN on graph for the analysis task of irregular spatial data. In this study, we used the classification task of building patterns as a case study to test this method, and experiments showed that this method has achieved outstanding results in identifying regular and irregular patterns, and has significantly improved in comparing with other methods.
We present an analysis into the inner workings of Convolutional Neural Networks (CNNs) for processing text. CNNs used for computer vision can be interpreted by projecting filters into image space, but for discrete sequence inputs CNNs remain a mystery. We aim to understand the method by which the networks process and classify text. We examine common hypotheses to this problem: that filters, accompanied by global max-pooling, serve as ngram detectors. We show that filters may capture several different semantic classes of ngrams by using different activation patterns, and that global max-pooling induces behavior which separates important ngrams from the rest. Finally, we show practical use cases derived from our findings in the form of model interpretability (explaining a trained model by deriving a concrete identity for each filter, bridging the gap between visualization tools in vision tasks and NLP) and prediction interpretability (explaining predictions). Code implementation is available online at github.com/sayaendo/interpreting-cnn-for-text.
Determining the programming language of a source code file has been considered in the research community; it has been shown that Machine Learning (ML) and Natural Language Processing (NLP) algorithms can be effective in identifying the programming language of source code files. However, determining the programming language of a code snippet or a few lines of source code is still a challenging task. Online forums such as Stack Overflow and code repositories such as GitHub contain a large number of code snippets. In this paper, we describe Source Code Classification (SCC), a classifier that can identify the programming language of code snippets written in 21 different programming languages. A Multinomial Naive Bayes (MNB) classifier is employed which is trained using Stack Overflow posts. It is shown to achieve an accuracy of 75% which is higher than that with Programming Languages Identification (PLI a proprietary online classifier of snippets) whose accuracy is only 55.5%. The average score for precision, recall and the F1 score with the proposed tool are 0.76, 0.75 and 0.75, respectively. In addition, it can distinguish between code snippets from a family of programming languages such as C, C++ and C#, and can also identify the programming language version such as C# 3.0, C# 4.0 and C# 5.0.
This paper presents the Axon AI's solution to the 2nd YouTube-8M Video Understanding Challenge, achieving the final global average precision (GAP) of 88.733% on the private test set (ranked 3rd among 394 teams, not considering the model size constraint), and 87.287% using a model that meets size requirement. Two sets of 7 individual models belonging to 3 different families were trained separately. Then, the inference results on a training data were aggregated from these multiple models and fed to train a compact model that meets the model size requirement. In order to further improve performance we explored and employed data over/sub-sampling in feature space, an additional regularization term during training exploiting label relationship, and learned weights for ensembling different individual models.
In many classification problems it is desirable to output well-calibrated probabilities on the different classes. We propose a robust, non-parametric method of calibrating probabilities called SplineCalib that utilizes smoothing splines to determine a calibration function. We demonstrate how applying certain transformations as part of the calibration process can improve performance on problems in deep learning and other domains where the scores tend to be "overconfident". We adapt the approach to multi-class problems and find that better calibration can improve accuracy as well as log-loss by better resolving uncertain cases. Finally, we present a cross-validated approach to calibration which conserves data. Significant improvements to log-loss and accuracy are shown on several different problems. We also introduce the ml-insights python package which contains an implementation of the SplineCalib algorithm.
Nowadays, modern Earth Observation systems continuously generate huge amounts of data. A notable example is represented by the Sentinel-2 mission, which provides images at high spatial resolution (up to 10m) with high temporal revisit period (every 5 days), which can be organized in Satellite Image Time Series (SITS). While the use of SITS has been proved to be beneficial in the context of Land Use/Land Cover (LULC) map generation, unfortunately, machine learning approaches commonly leveraged in remote sensing field fail to take advantage of spatio-temporal dependencies present in such data. Recently, new generation deep learning methods allowed to significantly advance research in this field. These approaches have generally focused on a single type of neural network, i.e., Convolutional Neural Networks (CNNs) or Recurrent Neural Networks (RNNs), which model different but complementary information: spatial autocorrelation (CNNs) and temporal dependencies (RNNs). In this work, we propose the first deep learning architecture for the analysis of SITS data, namely \method{} (DUal view Point deep Learning architecture for time series classificatiOn), that combines Convolutional and Recurrent neural networks to exploit their complementarity. Our hypothesis is that, since CNNs and RNNs capture different aspects of the data, a combination of both models would produce a more diverse and complete representation of the information for the underlying land cover classification task. Experiments carried out on two study sites characterized by different land cover characteristics (i.e., the \textit{Gard} site in France and the \textit{Reunion Island} in the Indian Ocean), demonstrate the significance of our proposal.
Toxic comment classification has become an active research field with many recently proposed approaches. However, while these approaches address some of the task's challenges others still remain unsolved and directions for further research are needed. To this end, we compare different deep learning and shallow approaches on a new, large comment dataset and propose an ensemble that outperforms all individual models. Further, we validate our findings on a second dataset. The results of the ensemble enable us to perform an extensive error analysis, which reveals open challenges for state-of-the-art methods and directions towards pending future research. These challenges include missing paradigmatic context and inconsistent dataset labels.
Capabilities to categorize a clause based on the type of situation entity (e.g., events, states and generic statements) the clause introduces to the discourse can benefit many NLP applications. Observing that the situation entity type of a clause depends on discourse functions the clause plays in a paragraph and the interpretation of discourse functions depends heavily on paragraph-wide contexts, we propose to build context-aware clause representations for predicting situation entity types of clauses. Specifically, we propose a hierarchical recurrent neural network model to read a whole paragraph at a time and jointly learn representations for all the clauses in the paragraph by extensively modeling context influences and inter-dependencies of clauses. Experimental results show that our model achieves the state-of-the-art performance for clause-level situation entity classification on the genre-rich MASC+Wiki corpus, which approaches human-level performance.
Thoracic diseases are very serious health problems that plague a large number of people. Chest X-ray is currently one of the most popular methods to diagnose thoracic diseases, playing an important role in the healthcare workflow. However, reading the chest X-ray images and giving an accurate diagnosis remain challenging tasks for expert radiologists. With the success of deep learning in computer vision, a growing number of deep neural network architectures were applied to chest X-ray image classification. However, most of the previous deep neural network classifiers were based on deterministic architectures which are usually very noise-sensitive and are likely to aggravate the overfitting issue. In this paper, to make a deep architecture more robust to noise and to reduce overfitting, we propose using deep generative classifiers to automatically diagnose thorax diseases from the chest X-ray images. Unlike the traditional deterministic classifier, a deep generative classifier has a distribution middle layer in the deep neural network. A sampling layer then draws a random sample from the distribution layer and input it to the following layer for classification. The classifier is generative because the class label is generated from samples of a related distribution. Through training the model with a certain amount of randomness, the deep generative classifiers are expected to be robust to noise and can reduce overfitting and then achieve good performances. We implemented our deep generative classifiers based on a number of well-known deterministic neural network architectures, and tested our models on the chest X-ray14 dataset. The results demonstrated the superiority of deep generative classifiers compared with the corresponding deep deterministic classifiers.
Genetic Programming (GP) is a computationally intensive technique which is naturally parallel in nature. Consequently, many attempts have been made to improve its run-time from exploiting highly parallel hardware such as GPUs. However, a second methodology of improving the speed of GP is through efficiency techniques such as subtree caching. However achieving parallel performance and efficiency is a difficult task. This paper will demonstrate an efficiency saving for GP compatible with the harnessing of parallel CPU hardware by exploiting tournament selection. Significant efficiency savings are demonstrated whilst retaining the capability of a high performance parallel implementation of GP. Indeed, a 74% improvement in the speed of GP is achieved with a peak rate of 96 billion GPop/s for classification type problems.
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.
Machine learning models are increasingly used to automate decisions that affect humans - deciding who should receive a loan, a job interview, or a social service. In such applications, a person should have the ability to change the decision of a model. When a person is denied a loan by a credit score, for example, they should be able to alter its input variables in a way that guarantees approval. Otherwise, they will be denied the loan as long as the model is deployed. More importantly, they will lack the ability to influence a decision that affects their livelihood. In this paper, we frame these issues in terms of recourse, which we define as the ability of a person to change the decision of a model by altering actionable input variables (e.g., income vs. age or marital status). We present integer programming tools to ensure recourse in linear classification problems without interfering in model development. We demonstrate how our tools can inform stakeholders through experiments on credit scoring problems. Our results show that recourse can be significantly affected by standard practices in model development, and motivate the need to evaluate recourse in practice.
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.