This article is part of Demystifying AI, a series of posts that (try) to disambiguate the jargon and myths surrounding AI. If you're living on this planet, you probably hear a lot about Artificial Intelligence these days. It's conquering every industry and domain, performing tasks more efficiently than humans, will put humans out of work, and may one day force humans into slavery. You might have also heard about narrow, general and super artificial intelligence, or about machine learning, deep learning, reinforced learning, supervised and unsupervised learning, neural networks, Bayesian networks and a whole lot of other confusing terms. That's a lot of jargon to cover in one post, and we'll leave the learning stuff for another day.

Suppose, we are given a set of $n$ elements to be clustered into $k$ (unknown) clusters, and an oracle/expert labeler that can interactively answer pair-wise queries of the form, "do two elements $u$ and $v$ belong to the same cluster?". The goal is to recover the optimum clustering by asking the minimum number of queries. In this paper, we initiate a rigorous theoretical study of this basic problem of query complexity of interactive clustering, and provide strong information theoretic lower bounds, as well as nearly matching upper bounds. Most clustering problems come with a similarity matrix, which is used by an automated process to cluster similar points together. Our main contribution in this paper is to show the dramatic power of side information aka similarity matrix on reducing the query complexity of clustering. A similarity matrix represents noisy pair-wise relationships such as one computed by some function on attributes of the elements. A natural noisy model is where similarity values are drawn independently from some arbitrary probability distribution $f_+$ when the underlying pair of elements belong to the same cluster, and from some $f_-$ otherwise. We show that given such a similarity matrix, the query complexity reduces drastically from $\Theta(nk)$ (no similarity matrix) to $O(\frac{k^2\log{n}}{\cH^2(f_+\|f_-)})$ where $\cH^2$ denotes the squared Hellinger divergence. Moreover, this is also information-theoretic optimal within an $O(\log{n})$ factor. Our algorithms are all efficient, and parameter free, i.e., they work without any knowledge of $k, f_+$ and $f_-$, and only depend logarithmically with $n$. Along the way, our work also reveals intriguing connection to popular community detection models such as the {\em stochastic block model}, significantly generalizes them, and opens up many venues for interesting future research.

Predictive models ground many state-of-the-art developments in statistical brain image analysis: decoding, MVPA, searchlight, or extraction of biomarkers. The principled approach to establish their validity and usefulness is cross-validation, testing prediction on unseen data. Here, I would like to raise awareness on error bars of cross-validation, which are often underestimated. Simple experiments show that sample sizes of many neuroimaging studies inherently lead to large error bars, eg $\pm$10% for 100 samples. The standard error across folds strongly underestimates them. These large error bars compromise the reliability of conclusions drawn with predictive models, such as biomarkers or methods developments where, unlike with cognitive neuroimaging MVPA approaches, more samples cannot be acquired by repeating the experiment across many subjects. Solutions to increase sample size must be investigated, tackling possible increases in heterogeneity of the data.

Data mining techniques are intended to extract information from large volumes of data (Tan et al., 2006). Data mining includes tasks such as classification, regression, clustering, or anomaly detection, among others. Traditional data mining techniques are typically applied to tabulated data. Novel techniques have also been devised for semi-structured or structured data, since exploiting the relationships among instances from a dataset leads to new research and development opportunities (Getoor and Diehl, 2005). For example, network data mining has been used to predict previously unknown protein interactions in protein-protein interaction networks (Martínez et al., 2014). It has also been used to study and predict future author collaborations and tendencies in co-authorship networks (Pavlov and Ichise, 2007). Different network mining techniques are used by popular internet search engines to rank the most relevant websites (Page et al., 1999). These are only some examples of the large number of applications of network data mining. There are many software tools that facilitate the analysis of networked data.

I am incredibly proud and excited to present the very first public product of Peptone, the Database of Structural Propensities of Proteins. Database of Structural Propensities of Proteins (dSPP) is the world's first interactive repository of structural and dynamic features of proteins with seamless integration for leading Machine Learning frameworks, Keras and Tensorflow. IDPs are implicated in numerous debilitating human pathologies, including Alzheimer's, Parkinson's, prion diseases, molecular basis of cancer, HIV, HSV, HVC, ZIKVR, and many others. Structural interpretation of propensity score for MOAG-4 protein, extracted from dSPP database https://peptone.io/dspp/entry/dSPP27058_0. MOAG-4 is known to enhance the process of protein aggregation in animal brain models, thus accelerating an early onset of Parkinson's disease.

This robot will ensure that your work trousers and shirts remain wrinkle-free. And it promises to do other chores, as well. If you're an enlightened, modern sort of person, you feel that ironing duties should be split 50/50 between men and women. They believe the task of ironing should reside entirely with one member of the household: The domestic robot. With that in mind, they've developed a humanoid robot called TEO that has mastered the art of ensuring your shirts and trousers are wrinkle-free -- courtesy of some smart image recognition algorithms.

Like many artificial intelligence companies in Canada, PeopleAnalytics.ai was happy to see the federal government's launch of its Pan-Canadian Artificial lntelligence Strategy for research and talent as part of the federal budget this year. The $125-million that the Liberals are committing to the project, to be administered through the Canadian Institute For Advanced Research (CIFAR), is expected to help to attract and retain top academic talent in this country. With the market for AI-related ideas and products expected to reach $47-billion by 2020, according to CIFAR, the sector has already attracted major investment from Facebook and Google, among others. For PeopleAnalytics.ai, based out of Toronto's MaRS Discovery District, Canada is at a crossroads where it has the ability to define exactly how it wants to mould its focus on AI. Mark Chaikelson, below, vice-president of product for PeopleAnalytics.ai, says the success of the government's plan, particularly in the AI clusters in Montreal, Toronto-Waterloo and Edmonton, will come down to three things: capital, customers and talent.

The venture capital community was shocked last week with Element AI's fundraise. How can a Canadian startup founded in late 2016 raise a Series-A in June 2017 for a cool $102 million? Microsoft, NVIDIA, and Intel Capital were among the illustrious group of VCs who put in this sum into a round that makes almost all other Series-A rounds look paltry. When you peruse Element's webpage you see a simple HTML site that talks vaguely about AI-Strategy consulting, expert matching, and AI-as-a-Service (AIaaS). It leaves you scratching your head, wondering how this firm's ideas are earth shattering enough for such a rocket trajectory.

This is the continuation of my mini-series on sentiment analysis of movie reviews, which originally appeared on recurrentnull.wordpress.com. Last time, we had a look at how well classical bag-of-words models worked for classification of the Stanford collection of IMDB reviews. As it turned out, the "winner" was Logistic Regression, using both unigrams and bigrams for classification. The best classification accuracy obtained was .89 So, bag-of-words models may be surprisingly successful, but they are limited in what they can do.

Let us discuss the applicability of these number systems. The real numbers are used in every aspect of physics. All quantities, measurements, and lengths of physical objects or processes are given as real numbers. Although complex numbers were formulated by mathematicians to help solve equations (i is the solution to the equation x2 -1), physicists started using complex numbers to discuss waves in the middle of the 19th century. In the 20th century, complex numbers became fundamental for the study of quantum mechanics. By now, the role of complex numbers is very important in many different branches of physics. The quaternions show up in physics but are not a major player. The octonions, the sedenions, and the larger number systems rarely arise in the physics literature.