In many scientific settings there is a need for adaptive experimental design to guide the process of identifying regions of the search space that contain as many true positives as possible subject to a low rate of false discoveries (i.e. false alarms). Such regions of the search space could differ drastically from a predicted set that minimizes 0/1 error and accurate identification could require very different sampling strategies. Like active learning for binary classification, this experimental design cannot be optimally chosen a priori, but rather the data must be taken sequentially and adaptively. However, unlike classification with 0/1 error, collecting data adaptively to find a set with high true positive rate and low false discovery rate (FDR) is not as well understood. In this paper we provide the first provably sample efficient adaptive algorithm for this problem. Along the way we highlight connections between classification, combinatorial bandits, and FDR control making contributions to each.

Many internet applications are powered by machine learned models, which are usually trained on labeled datasets obtained through either implicit / explicit user feedback signals or human judgments. Since societal biases may be present in the generation of such datasets, it is possible for the trained models to be biased, thereby resulting in potential discrimination and harms for disadvantaged groups. Motivated by the need for understanding and addressing algorithmic bias in web-scale ML systems and the limitations of existing fairness toolkits, we present the LinkedIn Fairness Toolkit (LiFT), a framework for scalable computation of fairness metrics as part of large ML systems. We highlight the key requirements in deployed settings, and present the design of our fairness measurement system. We discuss the challenges encountered in incorporating fairness tools in practice and the lessons learned during deployment at LinkedIn. Finally, we provide open problems based on practical experience.

Although precision and recall are standard performance measures for anomaly detection, their statistical properties in sequential detection settings are poorly understood. In this work, we formalize a notion of precision and recall with temporal tolerance for point-based anomaly detection in sequential data. These measures are based on time-tolerant confusion matrices that may be used to compute time-tolerant variants of many other standard measures. However, care has to be taken to preserve interpretability. We perform a statistical simulation study to demonstrate that precision and recall may overestimate the performance of a detector, when computed with temporal tolerance. To alleviate this problem, we show how to obtain null distributions for the two measures to assess the statistical significance of reported results.

Classification tasks in machine learning involving more than two classes are known by the name of "multi-class classification". Performance indicators are very useful when the aim is to evaluate and compare different classification models or machine learning techniques. Many metrics come in handy to test the ability of a multi-class classifier. Those metrics turn out to be useful at different stage of the development process, e.g. comparing the performance of two different models or analysing the behaviour of the same model by tuning different parameters. In this white paper we review a list of the most promising multi-class metrics, we highlight their advantages and disadvantages and show their possible usages during the development of a classification model.

We propose to learn invariant representations, in the data domain, to achieve interpretability in algorithmic fairness. Invariance implies a selectivity for high level, relevant correlations w.r.t. class label annotations, and a robustness to irrelevant correlations with protected characteristics such as race or gender. We introduce a non-trivial setup in which the training set exhibits a strong bias such that class label annotations are irrelevant and spurious correlations cannot be distinguished. To address this problem, we introduce an adversarially trained model with a null-sampling procedure to produce invariant representations in the data domain. To enable disentanglement, a partially-labelled representative set is used. By placing the representations into the data domain, the changes made by the model are easily examinable by human auditors. We show the effectiveness of our method on both image and tabular datasets: Coloured MNIST, the CelebA and the Adult dataset.

In several medical decision-making problems, such as antibiotic prescription, laboratory testing can provide precise indications for how a patient will respond to different treatment options. This enables us to "fully observe" all potential treatment outcomes, but while present in historical data, these results are infeasible to produce in real-time at the point of the initial treatment decision. Moreover, treatment policies in these settings often need to trade off between multiple competing objectives, such as effectiveness of treatment and harmful side effects. We present, compare, and evaluate three approaches for learning individualized treatment policies in this setting: First, we consider two indirect approaches, which use predictive models of treatment response to construct policies optimal for different trade-offs between objectives. Second, we consider a direct approach that constructs such a set of policies without intermediate models of outcomes. Using a medical dataset of Urinary Tract Infection (UTI) patients, we show that all approaches learn policies that achieve strictly better performance on all outcomes than clinicians, while also trading off between different objectives. We demonstrate additional benefits of the direct approach, including flexibly incorporating other goals such as deferral to physicians on simple cases.

While $12.7 billion of this figure goes to another merchant when a customer is turned away, it must to be noticed that false declines "are also making for a less efficient digital economy". This is because "$7.6 billion of potential spending never came about as the shopper lost interest. In the same report, a senior industry executive pointed out that re-visiting risk appetite is vital. Also, a "lot of sins can be hidden in the name of #fraud prevention, because fraud teams aren't always incentivised to have a very rigorous statistical measure of false positives and false negatives". "Many companies just don't want to get on the MasterCard and Visa chargeback programmes, and that's the guiding principle.

Semantic clone detection is the process of finding program elements with similar or equal runtime behavior. For example, detecting the semantic equality between the recursive and iterative implementation of the factorial computation. Semantic clone detection is the de facto technical boundary of clone detectors. This boundary was tested over the last years with interesting new approaches. This work contributes a semantic clone detection approach that detects clones with 0% syntactic similarity. We present Semantic Clone Detection via Probabilistic Software Modeling (SCD-PSM) as a stable and precise solution to semantic clone detection. PSM builds a probabilistic model of a program that is capable of evaluating and generating runtime data. SCD-PSM leverages this model and its model elements to finding behaviorally equal model elements. This behavioral equality is then generalized to semantic equality of the original program elements. It uses the likelihood between model elements as a distance metric. Then, it employs the likelihood ratio significance test to decide whether this distance is significant, given a pre-specified and controllable false-positive rate. The output of SCD-PSM are pairs of program elements (i.e., methods), their distance, and a decision whether they are clones or not. SCD-PSM yields excellent results with a Matthews Correlation Coefficient greater 0.9. These results are obtained on classical semantic clone detection problems such as detecting recursive and iterative versions of an algorithm, but also on complex problems used in coding competitions.

Measles is extremely contagious and is one of the leading causes of vaccine-preventable illness and death in developing countries, claiming more than 100,000 lives each year. Measles was declared eliminated in the US in 2000 due to decades of successful vaccination for the measles. As a result, an increasing number of US healthcare professionals and the public have never seen the disease. Unfortunately, the Measles resurged in the US in 2019 with 1,282 confirmed cases. To assist in diagnosing measles, we collected more than 1300 images of a variety of skin conditions, with which we employed residual deep convolutional neural network to distinguish measles rash from other skin conditions, in an aim to create a phone application in the future. On our image dataset, our model reaches a classification accuracy of 95.2%, sensitivity of 81.7%, and specificity of 97.1%, indicating the model is effective in facilitating an accurate detection of measles to help contain measles outbreaks.

The data made available for analysis are becoming more and more complex along several directions: high dimensionality, number of examples and the amount of labels per example. This poses a variety of challenges for the existing machine learning methods: coping with dataset with a large number of examples that are described in a high-dimensional space and not all examples have labels provided. For example, when investigating the toxicity of chemical compounds there are a lot of compounds available, that can be described with information rich high-dimensional representations, but not all of the compounds have information on their toxicity. To address these challenges, we propose semi-supervised learning of feature ranking. The feature rankings are learned in the context of classification and regression as well as in the context of structured output prediction (multi-label classification, hierarchical multi-label classification and multi-target regression). To the best of our knowledge, this is the first work that treats the task of feature ranking within the semi-supervised structured output prediction context. More specifically, we propose two approaches that are based on tree ensembles and the Relief family of algorithms. The extensive evaluation across 38 benchmark datasets reveals the following: Random Forests perform the best for the classification-like tasks, while for the regression-like tasks Extra-PCTs perform the best, Random Forests are the most efficient method considering induction times across all tasks, and semi-supervised feature rankings outperform their supervised counterpart across a majority of the datasets from the different tasks.