A few weeks ago, I wrote about how and why I was learning Machine Learning, mainly through Andrew Ng's Coursera course. Machine Learning is built on prerequisites, so much so that learning by first principles seems overwhelming. Do you really need to spend a month learning linear algebra? You'll be okay if you have some math and programming experience. You really just have to be familiar with Sigma notation and be able to express it in a for loop. Sure, your assignments will take longer to complete and the first few times you see those giant equations your head will spin, but you can do this!

DB: from a technical standpoint, how did AI help build the interface? KH: the main visualization -- the map of over 750 fonts in one two-dimensional plane -- would not have been possible without AI. DB: what did you find most challenging about working on the project? KH: when it comes to technology-driven explorations such as this one, one of the core challenges is to balance letting the technology speak for itself with creative license to make design decisions that may obscure the technology, but lead to a better experience. DB: what surprising things did you learn about machine learning algorithms in the process?

Artificial intelligence may still sound like a distant future, but don't be fooled. Social media users are already living AI-fueled lives online, and studies show they're eager for marketers to take the technology even further. A recent Weber Shandwick survey found that many more people view AI's advent as a good thing rather than a bad one -- both for society and for themselves. Forty-five percent of respondents foresee a positive societal impact; just seven percent see a negative one. When it comes to their personal lives, 52% expect AI to benefit them; again, only seven percent said it'll make things worse.

We think of computers as emotionless automatons and artificial intelligence as stoic, zen-like programs, mirroring Mr. Spock, devoid of prejudice and unable to be swayed by emotion. A team of researchers at Princeton University's engineering school have proven otherwise, in a new study. They say that AI picks up our innate biases about sex and race, even when we ourselves may be unaware of them. The results of this study were published in the journal Science. This may not be too surprising after a Microsoft snafu in March last year, when a chatbot named Tat had to be taken off Twitter.

I am sure by now, you have heard the phrase that has been thrown around quite a lot by mostly, venture capitalists: "Artificial Intelligence (AI) is the new mobile." The reason why this phrase has been echoed in the tech industry is to emphasize that AI is not a short-lived fad, rather a revolution like mobile. More importantly, they seem to be right as in the last five years, giant tech companies have been pouring money into this technology. In fact, over 200 private companies using AI algorithms across different verticals have been acquired since 2012, with over 30 acquisitions taking place in Q1'17 alone. The acquisitions of AI startups are getting feisty, too.

Alzheimer's disease has an unusual distinction: It's the illness that Americans fear most -- more than cancer, stroke or heart disease. The rhetoric surrounding Alzheimer's reflects this. People "fade away" and are tragically "robbed of their identities" as this incurable condition progresses, we're told time and again. Yet, a sizable body of research suggests this Alzheimer's narrative is mistaken. It finds that people with Alzheimer's and other types of dementia retain a sense of self and have a positive quality of life, overall, until the illness's final stages. They're energized by meaningful activities and value opportunities to express themselves.

We study strictly proper scoring rules in the Reproducing Kernel Hilbert Space. We propose a general Kernel Scoring rule and associated Kernel Divergence. We consider conditions under which the Kernel Score is strictly proper. We then demonstrate that the Kernel Score includes the Maximum Mean Discrepancy as a special case. We also consider the connections between the Kernel Score and the minimum risk of a proper loss function. We show that the Kernel Score incorporates more information pertaining to the projected embedded distributions compared to the Maximum Mean Discrepancy. Finally, we show how to integrate the information provided from different Kernel Divergences, such as the proposed Bhattacharyya Kernel Divergence, using a one-class classifier for improved two-sample hypothesis testing results.

Matrix completion models are among the most common formulations of recommender systems. Recent works have showed a boost of performance of these techniques when introducing the pairwise relationships between users/items in the form of graphs, and imposing smoothness priors on these graphs. However, such techniques do not fully exploit the local stationarity structures of user/item graphs, and the number of parameters to learn is linear w.r.t. the number of users and items. We propose a novel approach to overcome these limitations by using geometric deep learning on graphs. Our matrix completion architecture combines graph convolutional neural networks and recurrent neural networks to learn meaningful statistical graph-structured patterns and the non-linear diffusion process that generates the known ratings. This neural network system requires a constant number of parameters independent of the matrix size. We apply our method on both synthetic and real datasets, showing that it outperforms state-of-the-art techniques.

In this work we perform outlier detection using ensembles of neural networks obtained by variational approximation of the posterior in a Bayesian neural network setting. The variational parameters are obtained by sampling from the true posterior by gradient descent. We show our outlier detection results are comparable to those obtained using other efficient ensembling methods.

Automatically assessing emotional valence in human speech has historically been a difficult task for machine learning algorithms. The subtle changes in the voice of the speaker that are indicative of positive or negative emotional states are often "overshadowed" by voice characteristics relating to emotional intensity or emotional activation. In this work we explore a representation learning approach that automatically derives discriminative representations of emotional speech. In particular, we investigate two machine learning strategies to improve classifier performance: (1) utilization of unlabeled data using a deep convolutional generative adversarial network (DCGAN), and (2) multitask learning. Within our extensive experiments we leverage a multitask annotated emotional corpus as well as a large unlabeled meeting corpus (around 100 hours). Our speaker-independent classification experiments show that in particular the use of unlabeled data in our investigations improves performance of the classifiers and both fully supervised baseline approaches are outperformed considerably. We improve the classification of emotional valence on a discrete 5-point scale to 43.88% and on a 3-point scale to 49.80%, which is competitive to state-of-the-art performance.