Machine learning is the science of discovering statistical dependencies in data, and the use of those dependencies to perform predictions. During the last decade, machine learning has made spectacular progress, surpassing human performance in complex tasks such as object recognition, car driving, and computer gaming. However, the central role of prediction in machine learning avoids progress towards general-purpose artificial intelligence. As one way forward, we argue that causal inference is a fundamental component of human intelligence, yet ignored by learning algorithms. Causal inference is the problem of uncovering the cause-effect relationships between the variables of a data generating system. Causal structures provide understanding about how these systems behave under changing, unseen environments. In turn, knowledge about these causal dynamics allows to answer "what if" questions, describing the potential responses of the system under hypothetical manipulations and interventions. Thus, understanding cause and effect is one step from machine learning towards machine reasoning and machine intelligence. But, currently available causal inference algorithms operate in specific regimes, and rely on assumptions that are difficult to verify in practice. This thesis advances the art of causal inference in three different ways. First, we develop a framework for the study of statistical dependence based on copulas and random features. Second, we build on this framework to interpret the problem of causal inference as the task of distribution classification, yielding a family of novel causal inference algorithms. Third, we discover causal structures in convolutional neural network features using our algorithms. The algorithms presented in this thesis are scalable, exhibit strong theoretical guarantees, and achieve state-of-the-art performance in a variety of real-world benchmarks.

According to a recent study, machine learning algorithms are expected to replace 25% of the jobs across the world, in the next 10 years. With the rapid growth of big data and availability of programming tools like Python and R –machine learning is gaining mainstream presence for data scientists. Machine learning applications are highly automated and self-modifying which continue to improve over time with minimal human intervention as they learn with more data. For instance, Netflix's recommendation algorithm learns more about the likes and dislikes of a viewer based on the shows every viewer watches. To address the complex nature of various real world data problems, specialized machine learning algorithms have been developed that solve these problems perfectly.

I did a webcast earlier today about Bayesian statistics. Some time in the next week, the video should be available from O'Reilly. In the meantime, you can see my slides here: And here's a transcript of what I said: Thanks everyone for joining me for this webcast. At the bottom of this slide you can see the URL for my slides, so you can follow along at home. I'm Allen Downey and I'm a professor at Olin College, which is a new engineering college right outside Boston. Our mission is to fix engineering education, and one of the ways I'm working on that is by teaching Bayesian statistics. Bayesian methods have been the victim of a 200 year smear campaign. If you are interested in the history and the people involved, I recommend this book, The Theory That Would Not Die.

Naive Bayes Nearest Neighbour (NBNN) is a simple and effective framework which addresses many of the pitfalls of K-Nearest Neighbour (KNN) classification. It has yielded competitive results on several computer vision benchmarks. Its central tenet is that during NN search, a query is not compared to every example in a database, ignoring class information. Instead, NN searches are performed within each class, generating a score per class. A key problem with NN techniques, including NBNN, is that they fail when the data representation does not capture perceptual (e.g.~class-based) similarity. NBNN circumvents this by using independent engineered descriptors (e.g.~SIFT). To extend its applicability outside of image-based domains, we propose to learn a metric which captures perceptual similarity. Similar to how Neighbourhood Components Analysis optimizes a differentiable form of KNN classification, we propose "Class Conditional" metric learning (CCML), which optimizes a soft form of the NBNN selection rule. Typical metric learning algorithms learn either a global or local metric. However, our proposed method can be adjusted to a particular level of locality by tuning a single parameter. An empirical evaluation on classification and retrieval tasks demonstrates that our proposed method clearly outperforms existing learned distance metrics across a variety of image and non-image datasets.

This post was first published on my Linkedin page and posted here as a contributed post. In the last few months, I have had several people contact me about their enthusiasm for venturing into the world of data science and using Machine Learning (ML) techniques to probe statistical regularities and build impeccable data-driven products. However, I've observed that some actually lack the necessary mathematical intuition and framework to get useful results. This is the main reason I decided to write this blog post. Recently, there has been an upsurge in the availability of many easy-to-use machine and deep learning packages such as scikit-learn, Weka, Tensorflow etc. Machine Learning theory is a field that intersects statistical, probabilistic, computer science and algorithmic aspects arising from learning iteratively from data and finding hidden insights which can be used to build intelligent applications. Despite the immense possibilities of Machine and Deep Learning, a thorough mathematical understanding of many of these techniques is necessary for a good grasp of the inner workings of the algorithms and getting good results.

A significant portion of existing variable selection methods are only applicable to linear parametric models. Despite the linearity and additivity assumption, variable selection in linear regression models has been popular since 1970; refer to Akaike information criterion [AIC; Akaike (1973)]; Bayesian information criterion [BIC; Schwarz et al (1978)] and Risk inflation criterion [RIC; Foster and George (1994)]. Popular classical sparse-regression methods such as Least absolute shrinkage operator [LASSO; Tibshirani (1996); Efron et al (2004)], and related penalization methods (Fan and Li, 2001; Zou and Hastie, 2005; Zou, 2006; Zhang, 2010) have gained popularity over the last decade due to their simplicity, computational scalability and efficiency in prediction when the underlying relation between the response and the predictors can be adequately described by parametric models. Bayesian methods (Mitchell and Beauchamp, 1988; George and McCulloch, 1993, 1997) with sparsity inducing priors offers greater applicability beyond parametric models and are a convenient alternative when the underlying goal is in inference and uncertainty quantification. However, there is still a limited amount of literature which seriously considers relaxing the linearity assumption, particularly when the dimension of the predictors is high. Moreover, when the focus is on learning the interactions between the variables, parametric models are often restrictive since they require very many parameters to capture the higher-order interaction terms. 2 Smoothing based non-additive nonparametric regression methods (Lafferty and Wasser-man, 2008; Wahba, 1990; Green and Silverman, 1993; Hastie and Tibshirani, 1990) can accommodate a wide range of relationships between predictors and response leading to excellent predictive performance.

Bayesian inference in the presence of an intractable likelihood function is computationally challenging. When following a Markov chain Monte Carlo (MCMC) approach to approximate the posterior distribution in this context, one typically either uses MCMC schemes which target the joint posterior of the parameters and some auxiliary latent variables or pseudo-marginal Metropolis-Hastings (MH) schemes which mimic a MH algorithm targeting the marginal posterior of the parameters by approximating unbiasedly the intractable likelihood. In scenarios where the parameters and auxiliary variables are strongly correlated under the posterior and/or this posterior is multimodal, Gibbs sampling or Hamiltonian Monte Carlo (HMC) will perform poorly and the pseudo-marginal MH algorithm, as any other MH scheme, will be inefficient for high dimensional parameters. We propose here an original MCMC algorithm, termed pseudo-marginal HMC, which approximates the HMC algorithm targeting the marginal posterior of the parameters. We demonstrate through experiments that pseudo-marginal HMC can outperform significantly both standard HMC and pseudo-marginal MH schemes.

In this paper, we consider the problem of estimating the underlying graph associated with an Ising model given a number of independent and identically distributed samples. We adopt an \emph{approximate recovery} criterion that allows for a number of missed edges or incorrectly-included edges, in contrast with the widely-studied exact recovery problem. Our main results provide information-theoretic lower bounds on the sample complexity for graph classes imposing constraints on the number of edges, maximal degree, and other properties. We identify a broad range of scenarios where, either up to constant factors or logarithmic factors, our lower bounds match the best known lower bounds for the exact recovery criterion, several of which are known to be tight or near-tight. Hence, in these cases, approximate recovery has a similar difficulty to exact recovery in the minimax sense. Our bounds are obtained via a modification of Fano's inequality for handling the approximate recovery criterion, along with suitably-designed ensembles of graphs that can broadly be classed into two categories: (i) Those containing graphs that contain several isolated edges or cliques and are thus difficult to distinguish from the empty graph; (ii) Those containing graphs for which certain groups of nodes are highly correlated, thus making it difficult to determine precisely which edges connect them. We support our theoretical results on these ensembles with numerical experiments.

Directed graphical models provide a useful framework for modeling causal or directional relationships for multivariate data. Prior work has largely focused on identifiability and search algorithms for directed acyclic graphical (DAG) models. In many applications, feedback naturally arises and directed graphical models that permit cycles occur. In this paper we address the issue of identifiability for general directed cyclic graphical (DCG) models satisfying the Markov assumption. In particular, in addition to the faithfulness assumption which has already been introduced for cyclic models, we introduce two new identifiability assumptions, one based on selecting the model with the fewest edges and the other based on selecting the DCG model that entails the maximum number of d-separation rules. We provide theoretical results comparing these assumptions which show that: (1) selecting models with the largest number of d-separation rules is strictly weaker than the faithfulness assumption; (2) unlike for DAG models, selecting models with the fewest edges does not necessarily result in a milder assumption than the faithfulness assumption. We also provide connections between our two new principles and minimality assumptions. We use our identifiability assumptions to develop search algorithms for small-scale DCG models. Our simulation study supports our theoretical results, showing that the algorithms based on our two new principles generally out-perform algorithms based on the faithfulness assumption in terms of selecting the true skeleton for DCG models.

A treatment regime is a function that maps individual patient information to a recommended treatment, hence explicitly incorporating the heterogeneity in need for treatment across individuals. Patient responses are dichotomous and can be predicted through an unknown relationship that depends on the patient information and the selected treatment. The goal is to find the treatments that lead to the best patient responses on average. Each experiment is expensive, forcing us to learn the most from each experiment. We adopt a Bayesian approach both to incorporate possible prior information and to update our treatment regime continuously as information accrues, with the potential to allow smaller yet more informative trials and for patients to receive better treatment. By formulating the problem as contextual bandits, we introduce a knowledge gradient policy to guide the treatment assignment by maximizing the expected value of information, for which an approximation method is used to overcome computational challenges. We provide a detailed study on how to make sequential medical decisions under uncertainty to reduce health care costs on a real world knee replacement dataset. We use clustering and LASSO to deal with the intrinsic sparsity in health datasets. We show experimentally that even though the problem is sparse, through careful selection of physicians (versus picking them at random), we can significantly improve the success rates.