Over this decade, an increasing number of processes is being automated through classification algorithms, being essential that these are robust and reliable if we are to trust key operations based on their output. State-of-the-art classifiers perform extraordinarily well on standard data, but they have been shown to be vulnerable to adversarial examples, data instances specifically targeted at fooling the algorithms (Comiter, 2019). As a fundamental hypothesis, algorithms rely on the use of independent and identically distributed (iid) data for both the training and test phases. However, security aspects in classification, which form part of the field of adversarial machine learning (AML), question such hypothesis due to the presence of adversaries ready to modify the data to obtain a benefit and, thus, making both distributions differ. Stemming from the pioneering work in adversarial classification (AC) in Dalvi et al. (2004), the paradigm used to model the confrontation between adversaries and classification systems has been game theory, see recent reviews in Biggio and Roli (2018) and Zhou et al. (2018). As an example, the most popular attacks, including the fast gradient sign method (FGSM) (Goodfellow et al., 2014b), may be viewed from a game-theoretic perspective. Similarly, two of the most promising defence techniques, adversarial training (AT) (Madry et al., 2018), which trains the defender model with attacked samples, and adversarial logit pairing (ALP) (Kannan et al., 2018), which encourages the logits of the model to be the same for both standard and adversarial inputs, may be framed in game theoretic terms. This perspective typically entails common knowledge hypothesis (Hargreaves-Heap and Varoufakis, 2004) which, from a fundamental point of view, are not sustainable in settings such as security, as adversaries try to hide and conceal information. Recent work (Naveiro et al., 2019) presented ACRA, a novel approach for AC based on Adversarial Risk

We present the viewpoint that optimization problems encountered in machine learning can often be interpreted as minimizing a convex functional over a function space, but with a non-convex constraint set introduced by model parameterization. This observation allows us to repose such problems via a suitable relaxation as convex optimization problems in the space of distributions over the training parameters. We derive some simple relationships between the distribution-space problem and the original problem, e.g. a distribution-space solution is at least as good as a solution in the original space. Moreover, we develop a numerical algorithm based on mixture distributions to perform approximate optimization directly in distribution space. Consistency of this approximation is established and the numerical efficacy of the proposed algorithm is illustrated on simple examples. In both theory and practice, this formulation provides an alternative approach to large-scale optimization in machine learning.

Most current studies on survey analysis and risk tolerance modelling lack professional knowledge and domain-specific models. Given the effectiveness of generative adversarial learning in cross-domain information, we design an Asymmetric cross-Domain Generative Adversarial Network (ADGAN) for domain scale inequality. ADGAN utilizes the information-sufficient domain to provide extra information to improve the representation learning on the information-insufficient domain via domain alignment. We provide data analysis and user model on two data sources: Consumer Consumption Information and Survey Information. We further test ADGAN on a real-world dataset with view embedding structures and show ADGAN can better deal with the class imbalance and unqualified data space than state-of-the-art, demonstrating the effectiveness of leveraging asymmetrical domain information.

With the rapid development of science and technologies, massive data can be collected at a large speed, especially in internet and financial fields. It is generally recognized that two major challenges in large-scale learning are estimation and inference due to large amount of computation. For statistical inference on massive data sets, Kleiner et al. (2014) proposed the bag of little bootstrap (BLB) to assess the quality of estimators. However, they used only a small number of random subsets, and partial observations from each subset. This implies less efficiency in application.

The severity of knee osteoarthritis is graded using the 5-point Kellgren-Lawrence (KL) scale where healthy knees are assigned grade 0, and the subsequent grades 1-4 represent increasing severity of the affliction. Although several methods have been proposed in recent years to develop models that can automatically predict the KL grade from a given radiograph, most models have been developed and evaluated on datasets not sourced from India. These models fail to perform well on the radiographs of Indian patients. In this paper, we propose a novel method using convolutional neural networks to automatically grade knee radiographs on the KL scale. Our method works in two connected stages: in the first stage, an object detection model segments individual knees from the rest of the image; in the second stage, a regression model automatically grades each knee separately on the KL scale. We train our model using the publicly available Osteoarthritis Initiative (OAI) dataset and demonstrate that fine-tuning the model before evaluating it on a dataset from a private hospital significantly improves the mean absolute error from 1.09 (95% CI: 1.03-1.15) to 0.28 (95% CI: 0.25-0.32). Additionally, we compare classification and regression models built for the same task and demonstrate that regression outperforms classification.

Time series are an interesting frontier for kernel-based methods, for the simple reason that there is no kernel designed to represent them and their unique characteristics in full generality. Existing sequential kernels ignore the time indices, with many assuming that the series must be regularly-spaced; some such kernels are not even psd. In this manuscript, we show that a "series kernel" that is general enough to represent irregularly-spaced multivariate time series may be built out of well-known "vector kernels". We also show that all series kernels constructed using our methodology are psd, and are thus widely applicable. We demonstrate this point by formulating a Gaussian process-based strategy - with our series kernel at its heart - to make predictions about test series when given a training set. We validate the strategy experimentally by estimating its generalisation error on multiple datasets and comparing it to relevant baselines. We also demonstrate that our series kernel may be used for the more traditional setting of time series classification, where its performance is broadly in line with alternative methods.

Multi-class classification with a very large number of classes, or extreme classification, is a challenging problem from both statistical and computational perspectives. Most of the classical approaches to multi-class classification, including one-vs-rest or multi-class support vector machines, require the exact estimation of the classifier's margin, at both the training and the prediction steps making them intractable in extreme classification scenarios. In this paper, we study the impact of computing an approximate margin using nearest neighbor (ANN) search structures combined with locality-sensitive hashing (LSH). This approximation allows to dramatically reduce both the training and the prediction time without a significant loss in performance. We theoretically prove that this approximation does not lead to a significant loss of the risk of the model and provide empirical evidence over five publicly available large scale datasets, showing that the proposed approach is highly competitive with respect to state-of-the-art approaches on time, memory and performance measures.

Python is one of the most commonly used programming languages by data scientists and machine learning engineers. Although there has been no universal study on the prevalence of machine learning algorithms in Python in machine learning, a 2019 GitHub analysis of public repositories tagged as "machine-learning" not surprisingly found that Python was the most common language used. Python outranked other languages commonly used in the data science community including R, Scala, and Julia. This is all to say that if you're interested in being a data scientist or a machine learning engineer, then understanding Python should be on your to-do list. It's important to remember though that employing machine learning techniques involves more than just coding for coding's sake.

Python is one of the most commonly used programming languages by data scientists and machine learning engineers. Although there has been no universal study on the prevalence of machine learning algorithms in Python in machine learning, a 2019 GitHub analysis of public repositories tagged as "machine-learning" not surprisingly found that Python was the most common language used. Python outranked other languages commonly used in the data science community including R, Scala, and Julia. This is all to say that if you're interested in being a data scientist or a machine learning engineer, then understanding Python should be on your to-do list. It's important to remember though that employing machine learning techniques involves more than just coding for coding's sake.

The power grid frequency is the central observable in power system control, as it measures the balance of electrical supply and demand. A reliable frequency forecast can facilitate rapid control actions and may thus greatly improve power system stability. Here, we develop a weighted-nearest-neighbor (WNN) predictor to investigate how predictable the frequency trajectories are. Our forecasts for up to one hour are more precise than averaged daily profiles and could increase the efficiency of frequency control actions. Furthermore, we gain an increased understanding of the specific properties of different synchronous areas by interpreting the optimal prediction parameters (number of nearest neighbors, the prediction horizon, etc.) in terms of the physical system. Finally, prediction errors indicate the occurrence of exceptional external perturbations. Overall, we provide a diagnostics tool and an accurate predictor of the power grid frequency time series, allowing better understanding of the underlying dynamics.