In this paper, we analyze a fast parallel algorithm to efficiently select and build a set of $k$ random variables from a large set of $n$ candidate elements. This combinatorial optimization problem can be viewed in the context of feature selection for the prediction of a response variable. Using the adaptive sampling technique, which has recently been shown to exponentially speed up submodular maximization algorithms, we propose a new parallelizable algorithm that dramatically speeds up previous selection algorithms by reducing the number of rounds from $\mathcal O(k)$ to $\mathcal O(\log n)$ for objectives that do not conform to the submodularity property. We introduce a new metric to quantify the closeness of the objective function to submodularity and analyze the performance of adaptive sampling under this regime. We also conduct experiments on synthetic and real datasets and show that the empirical performance of adaptive sampling on not-submodular objectives greatly outperforms its theoretical lower bound. Additionally, the empirical running time drastically improved in all experiments without comprising the terminal value, showing the practicality of adaptive sampling.

A more formal discussion explanatory variables is of high practical importance is provided in Section 2. in many disciplines. Recent work exploits stability of regression coefficients or invariance However most state of the art methods suffer from scalability properties of models across different experimental problems since they scan all potential subsets of variables conditions for reconstructing the full causal and test whether the conditional distribution of Y given graph. These approaches generally do not scale a subset of variables is invariant across all environments well with the number of the explanatory variables (Peters et al., 2016) . This search is hence exponential in and are difficult to extend to nonlinear relationships. the number of covariates; the methods, while maintaining Contrary to existing work, we propose an appealing theoretical guarantees, are thus already computationally approach which even works for observational data hard for graphs of ten variables, and get infeasible alone, while still offering theoretical guarantees for larger graphs, unless one resorts to heuristic procedures.

This paper presents a novel data-driven approach for predicting the number of vegetation-related outages that occur in power distribution systems on a monthly basis. In order to develop an approach that is able to successfully fulfill this objective, there are two main challenges that ought to be addressed. The first challenge is to define the extent of the target area. An unsupervised machine learning approach is proposed to overcome this difficulty. The second challenge is to correctly identify the main causes of vegetation-related outages and to thoroughly investigate their nature. In this paper, these outages are categorized into two main groups: growth-related and weather-related outages, and two types of models, namely time series and non-linear machine learning regression models are proposed to conduct the prediction tasks, respectively. Moreover, various features that can explain the variability in vegetation-related outages are engineered and employed. Actual outage data, obtained from a major utility in the U.S., in addition to different types of weather and geographical data are utilized to build the proposed approach. Finally, by utilizing various time series models and machine learning methods, a comprehensive case study is carried out to demonstrate how the proposed approach can be used to successfully predict the number of vegetation-related outages and to help decision-makers to detect vulnerable zones in their systems.

While discriminative classifiers often yield strong predictive performance, missing feature values at prediction time can still be a challenge. Classifiers may not behave as expected under certain ways of substituting the missing values, since they inherently make assumptions about the data distribution they were trained on. In this paper, we propose a novel framework that classifies examples with missing features by computing the expected prediction on a given feature distribution. We then use geometric programming to learn a naive Bayes distribution that embeds a given logistic regression classifier and can efficiently take its expected predictions. Empirical evaluations show that our model achieves the same performance as the logistic regression with all features observed, and outperforms standard imputation techniques when features go missing during prediction time. Furthermore, we demonstrate that our method can be used to generate 'sufficient explanations' of logistic regression classifications, by removing features that do not affect the classification.

Regression problems that have closed-form solutions are well understood and can be easily implemented when the dataset is small enough to be all loaded into the RAM. Challenges arise when data is too big to be stored in RAM to compute the closed form solutions. Many techniques were proposed to overcome or alleviate the memory barrier problem but the solutions are often local optimal. In addition, most approaches require accessing the raw data again when updating the models. Parallel computing clusters are also expected if multiple models need to be computed simultaneously. We propose multiple learning approaches that utilize an array of sufficient statistics (SS) to address this big data challenge. This memory oblivious approach breaks the memory barrier when computing regressions with closed-form solutions, including but not limited to linear regression, weighted linear regression, linear regression with Box-Cox transformation (Box-Cox regression) and ridge regression models. The computation and update of the SS array can be handled at per row level or per mini-batch level. And updating a model is as easy as matrix addition and subtraction. Furthermore, multiple SS arrays for different models can be easily computed simultaneously to obtain multiple models at one pass through the dataset. We implemented our approaches on Spark and evaluated over the simulated datasets. Results showed our approaches can achieve closed-form solutions of multiple models at the cost of half training time of the traditional methods for a single model.

Decision trees and logistic regression are one of the most popular and well-known machine learning algorithms, frequently used to solve a variety of real-world problems. Stability of learning algorithms is a powerful tool to analyze their performance and sensitivity and subsequently allow researchers to draw reliable conclusions. The stability of these two algorithms has remained obscure. To that end, in this paper, we derive two stability notions for decision trees and logistic regression: hypothesis and pointwise hypothesis stability. Additionally, we derive these notions for L2-regularized logistic regression and confirm existing findings that it is uniformly stable. We show that the stability of decision trees depends on the number of leaves in the tree, i.e., its depth, while for logistic regression, it depends on the smallest eigenvalue of the Hessian matrix of the cross-entropy loss. We show that logistic regression is not a stable learning algorithm. We construct the upper bounds on the generalization error of all three algorithms. Moreover, we present a novel stability measuring framework that allows one to measure the aforementioned notions of stability. The measures are equivalent to estimates of expected loss differences at an input example and then leverage bootstrap sampling to yield statistically reliable estimates. Finally, we apply this framework to the three algorithms analyzed in this paper to confirm our theoretical findings and, in addition, we discuss the possibilities of developing new training techniques to optimize the stability of logistic regression, and hence decrease its generalization error.

Differential privacy is becoming a standard notion for performing privacy-preserving machine learning over sensitive data. It provides formal guarantees, in terms of the privacy budget, $\epsilon$, on how much information about individual training records is leaked by the model. While the privacy budget is directly correlated to the privacy leakage, the calibration of the privacy budget is not well understood. As a result, many existing works on privacy-preserving machine learning select large values of $\epsilon$ in order to get acceptable utility of the model, with little understanding of the concrete impact of such choices on meaningful privacy. Moreover, in scenarios where iterative learning procedures are used which require privacy guarantees for each iteration, relaxed definitions of differential privacy are often used which further tradeoff privacy for better utility. In this paper, we evaluate the impacts of these choices on privacy in experiments with logistic regression and neural network models. We quantify the privacy leakage in terms of advantage of the adversary performing inference attacks and by analyzing the number of members at risk for exposure. Our main findings are that current mechanisms for differential privacy for machine learning rarely offer acceptable utility-privacy tradeoffs: settings that provide limited accuracy loss provide little effective privacy, and settings that provide strong privacy result in useless models. Open source code is available at https://github.com/bargavj/EvaluatingDPML.

While machine learning (ML) methods have received a lot of attention in recent years, these methods are primarily for prediction. Empirical researchers conducting policy evaluations are, on the other hand, pre-occupied with causal problems, trying to answer counterfactual questions: what would have happened in the absence of a policy? Because these counterfactuals can never be directly observed (described as the "fundamental problem of causal inference") prediction tools from the ML literature cannot be readily used for causal inference. In the last decade, major innovations have taken place incorporating supervised ML tools into estimators for causal parameters such as the average treatment effect (ATE). This holds the promise of attenuating model misspecification issues, and increasing of transparency in model selection. One particularly mature strand of the literature include approaches that incorporate supervised ML approaches in the estimation of the ATE of a binary treatment, under the \textit{unconfoundedness} and positivity assumptions (also known as exchangeability and overlap assumptions). This article reviews popular supervised machine learning algorithms, including the Super Learner. Then, some specific uses of machine learning for treatment effect estimation are introduced and illustrated, namely (1) to create balance among treated and control groups, (2) to estimate so-called nuisance models (e.g. the propensity score, or conditional expectations of the outcome) in semi-parametric estimators that target causal parameters (e.g. targeted maximum likelihood estimation or the double ML estimator), and (3) the use of machine learning for variable selection in situations with a high number of covariates.

We investigate the complexity of logistic regression models which is defined by counting the number of indistinguishable distributions that the model can represent (Balasubramanian, 1997). We find that the complexity of logistic models with binary inputs does not only depend on the number of parameters but also on the distribution of inputs in a non-trivial way which standard treatments of complexity do not address. In particular, we observe that correlations among inputs induce effective dependencies among parameters thus constraining the model and, consequently, reducing its complexity. We derive simple relations for the upper and lower bounds of the complexity. Furthermore, we show analytically that, defining the model parameters on a finite support rather than the entire axis, decreases the complexity in a manner that critically depends on the size of the domain. Based on our findings, we propose a novel model selection criterion which takes into account the entropy of the input distribution. We test our proposal on the problem of selecting the input variables of a logistic regression model in a Bayesian Model Selection framework. In our numerical tests, we find that, while the reconstruction errors of standard model selection approaches (AIC, BIC, $\ell_1$ regularization) strongly depend on the sparsity of the ground truth, the reconstruction error of our method is always close to the minimum in all conditions of sparsity, data size and strength of input correlations. Finally, we observe that, when considering categorical instead of binary inputs, in a simple and mathematically tractable case, the contribution of the alphabet size to the complexity is very small compared to that of parameter space dimension. We further explore the issue by analysing the dataset of the "13 keys to the White House" which is a method for forecasting the outcomes of US presidential elections.

Complex black-box predictive models may have high accuracy, but opacity causes problems like lack of trust, lack of stability, sensitivity to concept drift. On the other hand, interpretable models require more work related to feature engineering, which is very time consuming. Can we train interpretable and accurate models, without timeless feature engineering? In this article, we show a method that uses elastic black-boxes as surrogate models to create a simpler, less opaque, yet still accurate and interpretable glass-box models. New models are created on newly engineered features extracted/learned with the help of a surrogate model. We show applications of this method for model level explanations and possible extensions for instance level explanations. We also present an example implementation in Python and benchmark this method on a number of tabular data sets.