This paper proposes a novel optimization principle and its implementation for unsupervised anomaly detection in sound (ADS) using an autoencoder (AE). The goal of unsupervised-ADS is to detect unknown anomalous sound without training data of anomalous sound. Use of an AE as a normal model is a state-of-the-art technique for unsupervised-ADS. To decrease the false positive rate (FPR), the AE is trained to minimize the reconstruction error of normal sounds and the anomaly score is calculated as the reconstruction error of the observed sound. Unfortunately, since this training procedure does not take into account the anomaly score for anomalous sounds, the true positive rate (TPR) does not necessarily increase. In this study, we define an objective function based on the Neyman-Pearson lemma by considering ADS as a statistical hypothesis test. The proposed objective function trains the AE to maximize the TPR under an arbitrary low FPR condition. To calculate the TPR in the objective function, we consider that the set of anomalous sounds is the complementary set of normal sounds and simulate anomalous sounds by using a rejection sampling algorithm. Through experiments using synthetic data, we found that the proposed method improved the performance measures of ADS under low FPR conditions. In addition, we confirmed that the proposed method could detect anomalous sounds in real environments.

Abstract--Violence is an epidemic in Brazil and a problem on the rise worldwide. Mobile devices provide communication technologies which can be used to monitor and alert about violent situations. However, current solutions, like panic buttons or safe words, might increase the loss of life in violent situations. We propose an embedded artificial intelligence solution, using natural language and speech processing technology, to silently alert someone who can help in this situation. The corpus used contains 400 positive phrases and 800 negative phrases, totaling 1,200 sentences which are classified using two well-known extraction methods for natural language processing tasks: bag-of-words and word embeddings and classified with a support vector machine. We describe the proof-of-concept product in development with promising results, indicating a path towards a commercial product. More importantly we show that model improvements via word embeddings and data augmentation techniques provide an intrinsically robust model. The final embedded solution also has a small footprint of less than 10 MB.

Interpretable classification models are built with the purpose of providing a comprehensible description of the decision logic to an external oversight agent. When considered in isolation, a decision tree, a set of classification rules, or a linear model, are widely recognized as human-interpretable. However, such models are generated as part of a larger analytical process, which, in particular, comprises data collection and filtering. Selection bias in data collection or in data pre-processing may affect the model learned. Although model induction algorithms are designed to learn to generalize, they pursue optimization of predictive accuracy. It remains unclear how interpretability is instead impacted. We conduct an experimental analysis to investigate whether interpretable models are able to cope with data selection bias as far as interpretability is concerned.

Classification methods based on sparse estimation have drawn much attention recently, due to their effectiveness in processing high-dimensional data such as images. In this paper, a method to improve the performance of a sparse representation classification (SRC) approach is proposed; it is then applied to the problem of online process monitoring of human workers, specifically manual material handling (MMH) operations monitored using wearable sensors (involving 111 sensor channels). Our proposed method optimizes the design matrix (aka dictionary) in the linear model used for SRC, minimizing its ill-posedness to achieve a sparse solution. This procedure is based on the idea of dictionary learning (DL): we optimize the design matrix formed by training datasets to minimize both redundancy and coherency as well as reducing the size of these datasets. Use of such optimized training data can subsequently improve classification accuracy and help decrease the computational time needed for the SRC; it is thus more applicable for online process monitoring. Performance of the proposed methodology is demonstrated using wearable sensor data obtained from manual material handling experiments, and is found to be superior to those of benchmark methods in terms of accuracy, while also requiring computational time appropriate for MMH online monitoring.

There are so many confusing and sometimes even counter-intuitive concepts in statistics. I mean, come on…even explaining the differences between Null Hypothesis and Alternative Hypothesis can be an ordeal. All I want to do is to understand and quantify the cost of my analytical models being wrong. For example, let's say that I'm a shepherd who has bad eyesight and have a hard time distinguishing between a wolf and a sheep dog. That's obviously a bad trait, because the costs of being wrong are very expensive: Okay, so I'm not a very good shepherd, but I am a very sophisticated shepherd and I've build a Neural Network application to distinguish a sheep dog from a wolf.

Abstract--We aim to predict and explain service failures in supply-chain networks, more precisely among last-mile pickup and delivery services to customers. We analyze a dataset of 500,000 services using (1) supervised classification with Random Forests, and (2) Association Rules. Our classifier reaches an average sensitivity of 0.7 and an average specificity of 0.7 for the 5 studied types of failure. Association Rules reassert the importance of confirmation calls to prevent failures due to customers not at home, show the importance of the time window size, slack time, and geographical location of the customer for the other failure types, and highlight the effect of the retailer company on several failure types. To reduce the occurrence of service failures, our data models could be coupled to optimizers, or used to define countermeasures to be taken by human dispatchers. Service failures are pervasive in supply-chain networks, with important consequences on their cost-efficiency and customer experience. We aim at predicting and explaining the cause of such failures, focusing on the last-mile pickup and delivery of items at customer locations. Such services are planned by optimizers solving some variations of the Vehicle-Routing Problem, in our case the Pickup and Delivery Problem with Time Windows (PDPTW [1]).

The last few years have seen an explosion of academic and popular interest in algorithmic fairness. Despite this interest and the volume and velocity of work that has been produced recently, the fundamental science of fairness in machine learning is still in a nascent state. In March 2018, we convened a group of experts as part of a CCC visioning workshop to assess the state of the field, and distill the most promising research directions going forward. This report summarizes the findings of that workshop. Along the way, it surveys recent theoretical work in the field and points towards promising directions for research.

Following the assumption that there may be nuanced features associated with some mammogram images that could lead to an unnecessary recall when interpreted by a radiologist, the researchers used a method based on convolutional neural networks (CNNs) to build a computer toolkit that could identify those images. The researchers trained CNN models using 14,860 images of 3,715 patients from the Full-Field Digital Mammography Dataset and the Digital Dataset of Screening Mammography. They investigated six classification scenarios that would help distinguish images of benign, malignant, and recalled-benign mammograms.

A central goal of algorithmic fairness is to reduce bias in automated decision making. An unavoidable tension exists between accuracy gains obtained by using sensitive information (e.g., gender or ethnic group) as part of a statistical model, and any commitment to protect these characteristics. Often, due to biases present in the data, using the sensitive information in the functional form of a classifier improves classification accuracy. In this paper we show how it is possible to get the best of both worlds: optimize model accuracy and fairness without explicitly using the sensitive feature in the functional form of the model, thereby treating different individuals equally. Our method is based on two key ideas. On the one hand, we propose to use Multitask Learning (MTL), enhanced with fairness constraints, to jointly learn group specific classifiers that leverage information between sensitive groups. On the other hand, since learning group specific models might not be permitted, we propose to first predict the sensitive features by any learning method and then to use the predicted sensitive feature to train MTL with fairness constraints. This enables us to tackle fairness with a three-pronged approach, that is, by increasing accuracy on each group, enforcing measures of fairness during training, and protecting sensitive information during testing. Experimental results on two real datasets support our proposal, showing substantial improvements in both accuracy and fairness.

The precision-recall (PR) curve has become the de facto replacement for the ROC curve in the presence of imbalance, namely where one class is far more likely than the other class. While the PR and ROC curves tend to be used interchangeably, they have some very different properties. Properties of the PR curve are the focus of this paper. We consider: (1) population PR curves, where complete distributional assumptions are specified for scores from both classes; and (2) empirical estimators of the PR curve, where we observe scores and no distributional assumptions are made. The properties have direct consequence on how the PR curve should, and should not, be used. For example, the empirical PR curve is not consistent when scores in the class of primary interest come from discrete distributions. On the other hand, a normal approximation can fit quite well for points on the empirical PR curve from continuously-defined scores, but convergence can be heavily influenced by the distributional setting, the amount of imbalance, and the point of interest on the PR curve.