One of the unresolved questions in the context of deep learning is the triumph of GD based optimization, which is guaranteed to converge to one of many local minima. To shed light on the nature of the solutions that are thus being discovered, we investigate the ensemble of solutions reached by the same network architecture, with different random initialization of weights and random mini-batches. Surprisingly, we observe that these solutions are in fact very similar - more often than not, each train and test example is either classified correctly by all the networks, or by none at all. Moreover, all the networks seem to share the same learning dynamics, whereby initially the same train and test examples are incorporated into the learnt model, followed by other examples which are learnt in roughly the same order. When different neural network architectures are compared, the same learning dynamics is observed even when one architecture is significantly stronger than the other and achieves higher accuracy. Finally, when investigating other methods that involve the gradual refinement of a solution, such as boosting, once again we see the same learning pattern. In all cases, it appears as if all the classifiers start by learning to classify correctly the same train and test examples, while the more powerful classifiers continue to learn to classify correctly additional examples. These results are incredibly robust, observed for a large variety of architectures, hyperparameters and different datasets of images. Thus we observe that different classification solutions may be discovered by different means, but typically they evolve in roughly the same manner and demonstrate a similar success and failure behavior. For a given dataset, such behavior seems to be strongly correlated with effective generalization, while the induced ranking of examples may reflect inherent structure in the data.

Depression is one of the most common mental health disorders, and a large number of depression people commit suicide each year. Potential depression sufferers do not consult psychological doctors because they feel ashamed or are unaware of any depression, which may result in severe delay of diagnosis and treatment. In the meantime, evidence shows that social media data provides valuable clues about physical and mental health conditions. In this paper, we argue that it is feasible to identify depression at an early stage by mining online social behaviours. Our approach, which is innovative to the practice of depression detection, does not rely on the extraction of numerous or complicated features to achieve accurate depression detection. Instead, we propose a novel classifier, namely, Inverse Boosting Pruning Trees (IBPT), which demonstrates a strong classification ability on a publicly accessible dataset with 7862 Twitter users. To comprehensively evaluate the classification capability of the IBPT, we use three real datasets from the UCI machine learning repository and the IBPT still obtains the best classification results against several state of the arts techniques. The results manifest that our proposed framework is promising for identifying social networks' users with depression.

A membership inference attack (MIA) against a machine learning model enables an attacker to determine whether a given data record was part of the model's training dataset or not. Such attacks have been shown to be practical both in centralized and federated settings, and pose a threat in many privacy-sensitive domains such as medicine or law enforcement. In the literature, the effectiveness of these attacks is invariably reported using metrics computed across the whole population. In this paper, we take a closer look at the attack's performance across different subgroups present in the data distributions. We introduce a framework that enables us to efficiently analyze the vulnerability of machine learning models to MIA. We discover that even if the accuracy of MIA looks no better than random guessing over the whole population, subgroups are subject to disparate vulnerability, i.e., certain subgroups can be significantly more vulnerable than others. We provide a theoretical definition for MIA vulnerability which we validate empirically both on synthetic and real data.

Real world observational data, together with causal inference, allow the estimation of causal effects when randomized controlled trials are not available. To be accepted into practice, such predictive models must be validated for the dataset at hand, and thus require a comprehensive evaluation toolkit, as introduced here. Since effect estimation cannot be evaluated directly, we turn to evaluating the various observable properties of causal inference, namely the observed outcome and treatment assignment. We developed a toolkit that expands established machine learning evaluation methods and adds several causal-specific ones. Evaluations can be applied in cross-validation, in a train-test scheme, or on the training data. Multiple causal inference methods are implemented within the toolkit in a way that allows modular use of the underlying machine learning models. Thus, the toolkit is agnostic to the machine learning model that is used. We showcase our approach using a rheumatoid arthritis cohort (consisting of about 120K patients) extracted from the IBM MarketScan(R) Research Database. We introduce an iterative pipeline of data definition, model definition, and model evaluation. Using this pipeline, we demonstrate how each of the evaluation components helps drive model selection and refinement of data extraction criteria in a way that provides more reproducible results and ensures that the causal question is answerable with available data. Furthermore, we show how the evaluation toolkit can be used to ensure that performance is maintained when applied to subsets of the data, thus allowing exploration of questions that move towards personalized medicine.

Neural networks, although a powerful engine in supervised methods, often require expensive tuning efforts for optimized performance. Habitability classes are hard to discriminate, especially when attributes used as hard markers of separation are removed from the data set. The solution is approached from the point of investigating analytical properties of the proposed activation functions. The theory of ordinary differential equations and fixed point are exploited to justify the "lack of tuning efforts" to achieve optimal performance compared to traditional activation functions. Additionally, the relationship between the proposed activation functions and the more popular ones is established through extensive analytical and empirical evidence. Finally, the activation functions have been implemented in plain vanilla feed-forward neural network to classify exoplanets.

We present a weakly-supervised data augmentation approach to improve Named Entity Recognition (NER) in a challenging domain: extracting biomedical entities (e.g., proteins) from the scientific literature. First, we train a neural NER (NNER) model over a small seed of fully-labeled examples. Second, we use a reference set of entity names (e.g., proteins in UniProt) to identify entity mentions with high precision, but low recall, on an unlabeled corpus. Third, we use the NNER model to assign weak labels to the corpus. Finally, we retrain our NNER model iteratively over the augmented training set, including the seed, the reference-set examples, and the weakly-labeled examples, which improves model performance. We show empirically that this augmented bootstrapping process significantly improves NER performance, and discuss the factors impacting the efficacy of the approach.

The increasing impact of algorithmic decisions on people's lives compels us to scrutinize their fairness and, in particular, the disparate impacts that ostensibly-color-blind algorithms can have on different groups. Examples include credit decisioning, hiring, advertising, criminal justice, personalized medicine, and targeted policymaking, where in some cases legislative or regulatory frameworks for fairness exist and define specific protected classes. In this paper we study a fundamental challenge to assessing disparate impacts in practice: protected class membership is often not observed in the data. This is particularly a problem in lending and healthcare. We consider the use of an auxiliary dataset, such as the US census, that includes class labels but not decisions or outcomes. We show that a variety of common disparity measures are generally unidentifiable aside for some unrealistic cases, providing a new perspective on the documented biases of popular proxy-based methods. We provide exact characterizations of the sharpest-possible partial identification set of disparities either under no assumptions or when we incorporate mild smoothness constraints. We further provide optimization-based algorithms for computing and visualizing these sets, which enables reliable and robust assessments -- an important tool when disparity assessment can have far-reaching policy implications. We demonstrate this in two case studies with real data: mortgage lending and personalized medicine dosing.

Recent years have seen a surge of interest in the problem of fair classification, which is concerned with disparities in classification output or performance when conditioned on a protected attribute such as race or gender, or ethnicity. Many measures of fairness have been introduced [1-14] and fairness-enhancing interventions have been proposed to mitigate these disparities [15]. Roughly categorized, these interventions either (i) change data used to train a classifier (pre-processing) [16-20], (ii) change a classifier's output (post-processing) [4, 21-24], or (iii) directly change a classification model to ensure fairness (in-processing) [5, 25-32]. This paper places more emphasis on probabilistic classification in which the outputs of interest are predicted probabilities of belonging to one of the classes, often referred to as scores, as opposed to binary predictions. Scores are desirable because they indicate confidences in predictions. We propose an optimization formulation for transforming scores to satisfy fairness constraints while minimizing the loss in utility. The formulation accommodates any fairness criteria that can be expressed as linear inequalities involving conditional means of scores, including variants of statistical parity (SP) [1] and equalized odds (EO) [4, 5]. We derive a closed-form expression for the optimal transformed scores and a convex dual optimization problem for the Lagrange multipliers that parametrize the transformation.

In this article, we investigate a family of classification algorithms defined by the principle of empirical risk minimization, in the high dimensional regime where the feature dimension $p$ and data number $n$ are both large and comparable. Based on recent advances in high dimensional statistics and random matrix theory, we provide under mixture data model a unified stochastic characterization of classifiers learned with different loss functions. Our results are instrumental to an in-depth understanding as well as practical improvements on this fundamental classification approach. As the main outcome, we demonstrate the existence of a universally optimal loss function which yields the best high dimensional performance at any given $n/p$ ratio.

Ensemble approaches for uncertainty estimation have recently been applied to the tasks of misclassification detection, out-of-distribution input detection and adversarial attack detection. Prior Networks have been proposed as an approach to efficiently emulating an ensemble of models by parameterising a Dirichlet prior distribution over output distributions. These models have been shown to outperform ensemble approaches, such as Monte-Carlo Dropout, on the task of out-of-distribution input detection. However, scaling Prior Networks to complex datasets with many classes is difficult using the training criteria originally proposed. This paper makes two contributions. Firstly, we show that the appropriate training criterion for Prior Networks is the reverse KL-divergence between Dirichlet distributions. Using this loss we successfully train Prior Networks on image classification datasets with up to 200 classes and improve out-of-distribution detection performance. Secondly, taking advantage of the new training criterion, this paper investigates using Prior Networks to detect adversarial attacks. It is shown that the construction of successful adaptive whitebox attacks, which affect the prediction and evade detection, against Prior Networks trained on CIFAR-10 and CIFAR-100 takes a greater amount of computational effort than against standard neural networks, adversarially trained neural networks and dropout-defended networks.