Active Learning for discriminative models has largely been studied with the focus on individual samples, with less emphasis on how classes are distributed or which classes are hard to deal with. In this work, we show that this is harmful. We propose a method based on the Bayes' rule, that can naturally incorporate class imbalance into the Active Learning framework. We derive that three terms should be considered together when estimating the probability of a classifier making a mistake for a given sample; i) probability of mislabelling a class, ii) likelihood of the data given a predicted class, and iii) the prior probability on the abundance of a predicted class. Implementing these terms requires a generative model and an intractable likelihood estimation. Therefore, we train a Variational Auto Encoder (VAE) for this purpose. To further tie the VAE with the classifier and facilitate VAE training, we use the classifiers' deep feature representations as input to the VAE. By considering all three probabilities, among them especially the data imbalance, we can substantially improve the potential of existing methods under limited data budget. We show that our method can be applied to classification tasks on multiple different datasets -- including one that is a real-world dataset with heavy data imbalance -- significantly outperforming the state of the art.

Survival models are used in various fields, such as the development of cancer treatment protocols. Although many statistical and machine learning models have been proposed to achieve accurate survival predictions, little attention has been paid to obtain well-calibrated uncertainty estimates associated with each prediction. The currently popular models are opaque and untrustworthy in that they often express high confidence even on those test cases that are not similar to the training samples, and even when their predictions are wrong. We propose a Bayesian framework for survival models that not only gives more accurate survival predictions but also quantifies the survival uncertainty better. Our approach is a novel combination of variational inference for uncertainty estimation, neural multi-task logistic regression for estimating nonlinear and time-varying risk models, and an additional sparsity-inducing prior to work with high dimensional data.

Machine learning models are prone to memorizing sensitive data, making them vulnerable to membership inference attacks in which an adversary aims to guess if an input sample was used to train the model. In this paper, we show that prior work on membership inference attacks may severely underestimate the privacy risks by relying solely on training custom neural network classifiers to perform attacks and focusing only on the aggregate results over data samples, such as the attack accuracy. To overcome these limitations, we first propose to benchmark membership inference privacy risks by improving existing non-neural network based inference attacks and proposing a new inference attack method based on a modification of prediction entropy. We also propose benchmarks for defense mechanisms by accounting for adaptive adversaries with knowledge of the defense and also accounting for the trade-off between model accuracy and privacy risks. Using our benchmark attacks, we demonstrate that existing defense approaches are not as effective as previously reported. Next, we introduce a new approach for fine-grained privacy analysis by formulating and deriving a new metric called the privacy risk score. Our privacy risk score metric measures an individual sample's likelihood of being a training member, which allows an adversary to perform membership inference attacks with high confidence. We experimentally validate the effectiveness of the privacy risk score metric and demonstrate that the distribution of the privacy risk score across individual samples is heterogeneous. Finally, we perform an in-depth investigation for understanding why certain samples have high privacy risk scores, including correlations with model sensitivity, generalization error, and feature embeddings. Our work emphasizes the importance of a systematic and rigorous evaluation of privacy risks of machine learning models.

One major impediment to the wider use of deep learning for clinical decision making is the difficulty of assigning a level of confidence to model predictions. Currently, deep Bayesian neural networks and sparse Gaussian processes are the main two scalable uncertainty estimation methods. However, deep Bayesian neural network suffers from lack of expressiveness, and more expressive models such as deep kernel learning, which is an extension of sparse Gaussian process, captures only the uncertainty from the higher level latent space. Therefore, the deep learning model under it lacks interpretability and ignores uncertainty from the raw data. In this paper, we merge features of the deep Bayesian learning framework with deep kernel learning to leverage the strengths of both methods for more comprehensive uncertainty estimation. Through a series of experiments on predicting the first incidence of heart failure, diabetes and depression applied to large-scale electronic medical records, we demonstrate that our method is better at capturing uncertainty than both Gaussian processes and deep Bayesian neural networks in terms of indicating data insufficiency and distinguishing true positive and false positive predictions, with a comparable generalisation performance. Furthermore, by assessing the accuracy and area under the receiver operating characteristic curve over the predictive probability, we show that our method is less susceptible to making overconfident predictions, especially for the minority class in imbalanced datasets. Finally, we demonstrate how uncertainty information derived by the model can inform risk factor analysis towards model interpretability.

Machine learning is driving development across many fields in science and engineering. A simple and efficient programming language could accelerate applications of machine learning in various fields. Currently, the programming languages most commonly used to develop machine learning algorithms include Python, MATLAB, and C/C ++. However, none of these languages well balance both efficiency and simplicity. The Julia language is a fast, easy-to-use, and open-source programming language that was originally designed for high-performance computing, which can well balance the efficiency and simplicity. This paper summarizes the related research work and developments in the application of the Julia language in machine learning. It first surveys the popular machine learning algorithms that are developed in the Julia language. Then, it investigates applications of the machine learning algorithms implemented with the Julia language. Finally, it discusses the open issues and the potential future directions that arise in the use of the Julia language in machine learning.

We consider anticipatory psychological models for human decision makers and their effect on sequential decision making. From a decision theoretic point of view, such models are time inconsistent meaning that Bellman's principle of optimality does not hold. The aim of this paper is to study how such an anxiety-based anticipatory utility can affect sequential decision making, such as quickest change detection, in multi-agent systems. We show that the interaction between anticipation-driven agents and sequential decision maker results in unusual (nonconvex) structure of the optimal decision policy. The methodology yields a useful mathematical framework for sensor interaction involving a human decision maker (with behavioral economics constraints) and a sensor equipped with automated sequential detector.

Deep neural networks (DNNs) show promise in breast cancer screening, but their robustness to input perturbations must be better understood before they can be clinically implemented. There exists extensive literature on this subject in the context of natural images that can potentially be built upon. However, it cannot be assumed that conclusions about robustness will transfer from natural images to mammogram images, due to significant differences between the two image modalities. In order to determine whether conclusions will transfer, we measure the sensitivity of a radiologist-level screening mammogram image classifier to four commonly studied input perturbations that natural image classifiers are sensitive to. We find that mammogram image classifiers are also sensitive to these perturbations, which suggests that we can build on the existing literature. We also perform a detailed analysis on the effects of low-pass filtering, and find that it degrades the visibility of clinically meaningful features called microcalcifications. Since low-pass filtering removes semantically meaningful information that is predictive of breast cancer, we argue that it is undesirable for mammogram image classifiers to be invariant to it. This is in contrast to natural images, where we do not want DNNs to be sensitive to low-pass filtering due to its tendency to remove information that is human-incomprehensible.

Deep Neural Networks (DNN) represent the state of the art in many tasks. However, due to their overparameterization, their generalization capabilities are in doubt and are still under study. Consequently, DNN can overfit and assign overconfident predictions, as they tend to learn highly oscillating decision thresholds. This has been shown to affect the calibration of the confidences assigned to unseen data. Data Augmentation (DA) strategies have been proposed to overcome some of these limitations. One of the most popular is Mixup, which has shown a great ability to improve the accuracy of these models. Recent work has provided evidence that Mixup also improves the uncertainty quantification and calibration of DNN. In this work, we argue and provide empirical evidence that, due to its fundamentals, Mixup does not necessarily improve calibration. Based on our observations we propose a new loss function that improves the calibration, and also sometimes the accuracy. Our loss is inspired by Bayes decision theory and introduces a new training framework for designing losses for probabilistic modelling. We provide state-of-the-art accuracy with consistent improvements in calibration performance.

Most of the daily news and recently published scientific papers on research, innovations, and applications in artificial intelligence (AI) refer to what is known as machine learning--algorithms using massive amounts of data and various methodologies to find patterns, support decisions, make predictions, or, for the deep learning part, self-identify important features in data. However, AI is a complex concept to grasp, and most people have little understanding of what it really is. AI was founded as an academic discipline in 1956 and, despite its youth, already has a rich history [1,2]. In more than 60 years of exploration and progress, AI has become a large field of research and development involving multidisciplinary approaches to address many challenges, from theoretical frameworks, methods, and tools to real implementations, risk analysis, and impact measures. The definition of AI is a moving target and changes over time with the evolution of the field. Since its early days, the field of AI has allowed the development of many techniques supporting decision support and prediction, as it is usually made by humans. As early as 1958, a perceptron was expected to be able "to walk, talk, see, write, reproduce itself and be conscious of its existence," which led a large scientific controversy between neural network and symbolic reasoning approaches [3].

We review the current literature concerned with information plane analyses of neural network classifiers. While the underlying information bottleneck theory and the claim that information-theoretic compression is causally linked to generalization are plausible, empirical evidence was found to be both supporting and conflicting. We review this evidence together with a detailed analysis how the respective information quantities were estimated. Our analysis suggests that compression visualized in information planes is not information-theoretic, but is rather compatible with geometric compression of the activations.