Say you have a collection of customers with a variety of characteristics such as age, location, and financial history, and you wish to discover patterns and sort them into clusters. Or perhaps you have a set of texts, such as wikipedia pages, and you wish to segment them into categories based on their content. This is the world of unsupervised learning, called as such because you are not guiding, or supervising, the pattern discovery by some prediction task, but instead uncovering hidden structure from unlabeled data. Unsupervised learning encompasses a variety of techniques in machine learning, from clustering to dimension reduction to matrix factorization. In this course, you'll learn the fundamentals of unsupervised learning and implement the essential algorithms using scikit-learn and scipy. You will learn how to cluster, transform, visualize, and extract insights from unlabeled datasets, and end the course by building a recommender system to recommend popular musical artists.

Chinese internet giant Baidu has released its fiscal 2016 results, reporting operating profit of 10.04 billion yuan, a 13.9 percent decrease from the year prior, on revenue of 70.5 billion yuan, up 6.3 percent from fiscal 2015. Mobile revenue represented 63 percent of total revenue in 2016, up from 2015's 53 percent, while online marketing revenues for the year reached 64.5 billion yuan, a 0.8 percent increase from 2015. Earnings before interest, taxation, depreciation, and amortisation (EBITDA) in fiscal 2016 was 15.8 billion yuan, down 5.2 percent from 2015, while operating cashflow stood at 22.2 billion yuan. In its full year and fourth quarter results, Baidu chair and CEO Robin Li said that the company is well-positioned to lead China's growth in artificial intelligence (AI), based on its technology, data, and talent pool. "Our existing platform, including our search and newsfeed products, are enhanced by AI and enriched by our content and services ecosystem," Li said.

This article was originally published on http://techemergence.com/ and it s the result of a collaboration between BootstrapLabs and Techemergence. Job automation predictions from an individual expert typically draw from years of academic research experience, or time "in the trenches" of industry. With growing interest and speculation on the job market of the next decade, we set out to garner a perspective as to what Silicon Valley thinks about the possibilities of automations in various business tasks. We wanted to know – what work functions have the most potential for near-term automation? In the infographics and article below, we explore the survey responses from nearly 80 Bay Area investors, founders, and tech folks – on which business functions have the greatest potential for automation today, and in the coming five years ahead.

Recent advances in machine learning have enabled a number of applications for health and well-being, marketing and social robots, among others. Traditional machine learning relies mainly on generic models: models tuned to an average target population. However, the'good' performance by these generic models doesn't necessarily translate to each individual in the group. While this can be acceptable in certain domains (e.g., marketing research), when it comes to, for instance, health and well-being, new systems need be optimized and work for each person. They should also help an individual to see, for example, which factors they might change in their life to improve their health or mood.

Recently, tech leaders including Tesla CEO Elon Musk and billionaire Bill Gates warned about robots taking over jobs. However, a new report shows people are not that concerned about job automation. The report The Robots Are Coming, But Not For Me, is based on an online survey of more than 2,000 adults commissioned by consumer engagements firm LivePerson last month. Although experts agree robots, artificial intelligence and automated technology will have at least some effect on global economies, the majority of people think they'll be alright but say other professionals will be affected by mass layoffs, the report found. Among those surveyed, a high 88 percent said they were not very worried about losing their job after they were told about an Oxford University study that said 47 percent of US jobs were "highly likely" to be replaced by technology.

Word embedding models offer continuous vector representations that can capture rich contextual semantics based on their word co-occurrence patterns. While these word vectors can provide very effective features used in many NLP tasks such as clustering similar words and inferring learning relationships, many challenges and open research questions remain. In this paper, we propose a solution that aligns variations of the same model (or different models) in a joint low-dimensional latent space leveraging carefully generated synthetic data points. This generative process is inspired by the observation that a variety of linguistic relationships is captured by simple linear operations in embedded space. We demonstrate that our approach can lead to substantial improvements in recovering embeddings of local neighborhoods.

Complex high-dimensional datasets pose a variety of computational and statistical challenges. In attempts to address these challenges, the past decade has witnessed a significant amount of research on sparsity constraints in statistical models. Sparsity constraints have practical and theoretical benefits: often they lead to more interpretable models, that can be estimated efficiently even in the high-dimensional regime where the sample size n can be dwarfed by the model dimension d. In addition to being convenient from a methodological and theoretical standpoint, sparse models have also had enormous practical impact, for instance in computational biology, neuroscience and applied machine learning. On the other hand, much of the theoretical literature on sparse estimation has focused on providing guarantees under strong, often impractical, generative assumptions.

We study the problem of using i.i.d. samples from an unknown multivariate probability distribution $p$ to estimate the mutual information of $p$. This problem has recently received attention in two settings: (1) where $p$ is assumed to be Gaussian and (2) where $p$ is assumed only to lie in a large nonparametric smoothness class. Estimators proposed for the Gaussian case converge in high dimensions when the Gaussian assumption holds, but are brittle, failing dramatically when $p$ is not Gaussian. Estimators proposed for the nonparametric case fail to converge with realistic sample sizes except in very low dimensions. As a result, there is a lack of robust mutual information estimators for many realistic data. To address this, we propose estimators for mutual information when $p$ is assumed to be a nonparanormal (a.k.a., Gaussian copula) model, a semiparametric compromise between Gaussian and nonparametric extremes. Using theoretical bounds and experiments, we show these estimators strike a practical balance between robustness and scaling with dimensionality.

For many internet businesses, presenting a given list of items in an order that maximizes a certain metric of interest (e.g., click-through-rate, average engagement time etc.) is crucial. We approach the aforementioned task from a learning-to-rank perspective which reveals a new problem setup. In traditional learning-to-rank literature, it is implicitly assumed that during the training data generation one has access to the \emph{best or desired} order for the given list of items. In this work, we consider a problem setup where we do not observe the desired ranking. We present two novel solutions: the first solution is an extension of already existing listwise learning-to-rank technique--Listwise maximum likelihood estimation (ListMLE)--while the second one is a generic machine learning based framework that tackles the problem in its entire generality. We discuss several challenges associated with this generic framework, and propose a simple \emph{item-payoff} and \emph{positional-gain} model that addresses these challenges. We provide training algorithms, inference procedures, and demonstrate the effectiveness of the two approaches over traditional ListMLE on synthetic as well as on real-life setting of ranking news articles for increased dwell time.

Many activation functions have been proposed in the past, but selecting an adequate one requires trial and error. We propose a new methodology of designing activation functions within a neural network at each layer. We call this technique an "activation ensemble" because it allows the use of multiple activation functions at each layer. This is done by introducing additional variables, $\alpha$, at each activation layer of a network to allow for multiple activation functions to be active at each neuron. By design, activations with larger $\alpha$ values at a neuron is equivalent to having the largest magnitude. Hence, those higher magnitude activations are "chosen" by the network. We implement the activation ensembles on a variety of datasets using an array of Feed Forward and Convolutional Neural Networks. By using the activation ensemble, we achieve superior results compared to traditional techniques. In addition, because of the flexibility of this methodology, we more deeply explore activation functions and the features that they capture.