We develop a family of reformulations of an arbitrary consistent linear system into a stochastic problem. The reformulations are governed by two user-defined parameters: a positive definite matrix defining a norm, and an arbitrary discrete or continuous distribution over random matrices. Our reformulation has several equivalent interpretations, allowing for researchers from various communities to leverage their domain specific insights. In particular, our reformulation can be equivalently seen as a stochastic optimization problem, stochastic linear system, stochastic fixed point problem and a probabilistic intersection problem. We prove sufficient, and necessary and sufficient conditions for the reformulation to be exact. Further, we propose and analyze three stochastic algorithms for solving the reformulated problem---basic, parallel and accelerated methods---with global linear convergence rates. The rates can be interpreted as condition numbers of a matrix which depends on the system matrix and on the reformulation parameters. This gives rise to a new phenomenon which we call stochastic preconditioning, and which refers to the problem of finding parameters (matrix and distribution) leading to a sufficiently small condition number. Our basic method can be equivalently interpreted as stochastic gradient descent, stochastic Newton method, stochastic proximal point method, stochastic fixed point method, and stochastic projection method, with fixed stepsize (relaxation parameter), applied to the reformulations.

We study the Stochastic Gradient Langevin Dynamics (SGLD) algorithm for non-convex optimization. The algorithm performs stochastic gradient descent, where in each step it injects appropriately scaled Gaussian noise to the update. We analyze the algorithm's hitting time to an arbitrary subset of the parameter space. Two results follow from our general theory: First, we prove that for empirical risk minimization, if the empirical risk is point-wise close to the (smooth) population risk, then the algorithm achieves an approximate local minimum of the population risk in polynomial time, escaping suboptimal local minima that only exist in the empirical risk. Second, we show that SGLD improves on one of the best known learnability results for learning linear classifiers under the zero-one loss.

In order to interact intelligently with objects in the world, animals must first transform neural population responses into estimates of the dynamic, unknown stimuli which caused them. The Bayesian solution to this problem is known as a Bayes filter, which applies Bayes' rule to combine population responses with the predictions of an internal model. In this paper we present a method for learning to approximate a Bayes filter when the stimulus dynamics are unknown. To do this we use the inferential properties of probabilistic population codes to compute Bayes' rule, and train a neural network to compute approximate predictions by the method of maximum likelihood. In particular, we perform stochastic gradient descent on the negative log-likelihood with a novel approximation of the gradient. We demonstrate our methods on a finite-state, a linear, and a nonlinear filtering problem, and show how the hidden layer of the neural network develops tuning curves which are consistent with findings in experimental neuroscience.

Multi-Task Learning (MTL) can enhance a classifier's generalization performance by learning multiple related tasks simultaneously. Conventional MTL works under the offline or batch setting, and suffers from expensive training cost and poor scalability. To address such inefficiency issues, online learning techniques have been applied to solve MTL problems. However, most existing algorithms of online MTL constrain task relatedness into a presumed structure via a single weight matrix, which is a strict restriction that does not always hold in practice. In this paper, we propose a robust online MTL framework that overcomes this restriction by decomposing the weight matrix into two components: the first one captures the low-rank common structure among tasks via a nuclear norm and the second one identifies the personalized patterns of outlier tasks via a group lasso. Theoretical analysis shows the proposed algorithm can achieve a sub-linear regret with respect to the best linear model in hindsight. Even though the above framework achieves good performance, the nuclear norm that simply adds all nonzero singular values together may not be a good low-rank approximation. To improve the results, we use a log-determinant function as a non-convex rank approximation. The gradient scheme is applied to optimize log-determinant function and can obtain a closed-form solution for this refined problem. Experimental results on a number of real-world applications verify the efficacy of our method.

There are three ways to look at data. This is when you look at data from the (potentially very recent) past. It allows you to explore the questions what happened and why did it happen? This is looking at things as they happen. In many cases, monitoring is used to find abnormalities.

That's a broad topic which has been treated many times. Much of what has been written on this topic is good, much is bad. But I find that the stats vs. machine learning argument, at that level, tends to focus on the forest at the cost of completely overlooking the trees. Shah's definitions, which I believe are reflective of many approaches, tend to focus on different ends of the respective spectrums of each of these concepts, treating machine learning as a practical activity and statistics as a theoretical abstraction (and, yes, I'm lumping "statistical modeling" together with "statistics" in this case... at least, for now). The relationship between statistics and machine learning is actually a highly complex one, and merely defining the 2 concepts is not helpful in dissecting this connection.

Any author would like to know if his/her article will be successful or not. Here is an attempt to deal with this task. We crawled 5000 URLs and for each URL we downloaded the title, body of the article and parameters: number of likes (not including Facebook likes), number of comments, number of views, article creation date and date of the last comment. First, we got rid of empty (or deleted), very short (less than 100 characters long) and "not found" articles, thus getting 2000 articles with associated parameters. Then we removed articles with missing parameters and ended up with only 1207 articles. Second, for every article we conducted tokenization of words.

This is a two-part series about using machine learning to hack my taste in music. In this first piece, I applied unsupervised learning techniques and tools on Pandora data to analyze songs that I like. The second part, which will be published soon, is about using supervised on Spotify data to predict whether or not I will like a song. If you take a look at my top tracks on Last.FM, you'll notice a smorgasbord of tracks from artists like LCD Soundsytem, Jimi Hendrix, and Kanye West. When I make a playlist, it's not uncommon for me to include some 80's post-disco, 2000s indie rock, and Nigerian or Turkish funk.

Explain what regularization is and why it is useful. What are the benefits and drawbacks of specific methods, such as ridge regression and LASSO? Explain what a local optimum is and why it is important in a specific context, such as k-means clustering. What are specific ways for determining if you have a local optimum problem? What can be done to avoid local optima?

In high-dimensional classification settings, we wish to seek a balance between high power and ensuring control over a desired loss function. In many settings, the points most likely to be misclassified are those who lie near the decision boundary of the given classification method. Often, these uninformative points should not be classified as they are noisy and do not exhibit strong signals. In this paper, we introduce the Thresholding Method to parameterize the problem of determining which points exhibit strong signals and should be classified. We demonstrate the empirical performance of this novel calibration method in providing loss function control at a desired level, as well as explore how the method assuages the effect of overfitting. We explore the benefits of error control through the Thresholding Method in difficult, high-dimensional, simulated settings. Finally, we show the flexibility of the Thresholding Method through applying the method in a variety of real data settings.