In many machine learning applications, submodular functions have been used as a model for evaluating the utility or payoff of a set such as news items to recommend, sensors to deploy in a terrain, nodes to influence in a social network, to name a few. At the heart of all these applications is the assumption that the underlying utility/payoff function is known a priori, hence maximizing it is in principle possible. In real life situations, however, the utility function is not fully known in advance and can only be estimated via interactions. For instance, whether a user likes a movie or not can be reliably evaluated only after it was shown to her. Or, the range of influence of a user in a social network can be estimated only after she is selected to advertise the product. We model such problems as an interactive submodular bandit optimization, where in each round we receive a context (e.g., previously selected movies) and have to choose an action (e.g., propose a new movie). We then receive a noisy feedback about the utility of the action (e.g., ratings) which we model as a submodular function over the context-action space. We develop SM-UCB that efficiently trades off exploration (collecting more data) and exploration (proposing a good action given gathered data) and achieves a $O(\sqrt{T})$ regret bound after $T$ rounds of interaction. Given a bounded-RKHS norm kernel over the context-action-payoff space that governs the smoothness of the utility function, SM-UCB keeps an upper-confidence bound on the payoff function that allows it to asymptotically achieve no-regret. Finally, we evaluate our results on four concrete applications, including movie recommendation (on the MovieLense data set), news recommendation (on Yahoo! Webscope dataset), interactive influence maximization (on a subset of the Facebook network), and personalized data summarization (on Reuters Corpus). In all these applications, we observe that SM-UCB consistently outperforms the prior art.

We prove near-tight concentration of measure for polynomial functions of the Ising model, under high temperature, improving the radius of concentration guaranteed by known results by polynomial factors in the dimension (i.e.~the number of nodes in the Ising model). We show that our results are optimal up to logarithmic factors in the dimension. We obtain our results by extending and strengthening the exchangeable-pairs approach used to prove concentration of measure in this setting by Chatterjee. We demonstrate the efficacy of such functions as statistics for testing the strength of interactions in social networks in both synthetic and real world data.

As we end 2017, I'm tired of writing "lecturing" blogs about what organizations should be doing to master data monetization in order to power their business models and achieve digital transformation. While the objective of every organization should be to master big data and data science (artificial intelligence, machine learning, deep learning) to drive "data monetization," let's take a breath and have some fun. My recent ankle surgery afforded me the opportunity to binge watch "Game of Thrones." As I watched the impending battle between the White Walkers and humanity, I couldn't help but identify a number of lessons that we can learn from Jon Snow's battle with the leader of the White Walkers... and the power of Valyrian steel! Game of Thrones and data, not exactly two things you think are in harmony, but this is where I find myself.

WHEN the first printed books with illustrations started to appear in the 1470s in the German city of Augsburg, wood engravers rose up in protest. Worried about their jobs, they literally stopped the presses. In fact, their skills turned out to be in higher demand than before: somebody had to illustrate the growing number of books. Fears about the impact of technology on jobs have resurfaced periodically ever since. The latest bout of anxiety concerns the arrival of artificial intelligence (AI).

No matter how many Amazon Echo commercials you see, it takes a little time to adjust to Alexa. Putting a virtual assistant in your home signals a change in lifestyle, sort of like adopting a puppy. There will be a lot of trial-and-error, but once you find your rhythm, you'll forget what life was like without her. The Amazon Echo listens for the wake word, "Alexa." But, frankly, I was shocked by how many conversations were recorded by my Echo that did not include the wake word. Click here to learn how to listen to everything Amazon Echo has ever heard.

"We should study Science Fiction in order to understand what someday could become Science Fact." This is the time of year when everyone makes his or her predictions for 2018. I have my predictions as well, but wanted to do something a bit more fun. So I thought I'd look backwards to the state of technology 50 years ago to gain some insights that we can use to make projections about 2018. That is, what "predictions" made in the 1950's might tell us about 2018.

If we use word embedding in our seq2seq model, why don't we just use the distance between 2 vectors as a loss function instead of softmax cross entropy?

For this episode, I have created another chatbot that is aimed at helping factory workers to intervene on machines whenever they encounter operating problems. This factory comes with a specific jargon and workers are surrounded by permanent noise which can obfuscate worker statements when they give vocal orders. We'll tackle these constraints by leveraging the Custom Speech service with the bot framework. We'll also see how Custom Speech differs from Speech Priming that I talked about in episode 8. If you're not yet familiar with the bot framework and the cognitive services in general, I strongly advise you to watch my other episodes as I will only focus on Custom Speech and I will not explain things I have already explained in the previous episodes.

Has any work been do on neural networks that can generate their own labels? The thought would be, humans can improve at games like chess without being told explicitly what good moves are. Someone can play lots of games and start to lable certain ideas as good or bad. Maybe they notice that having center control leads to more wins, and moving the same price many times leads to losses. With this in mind I am wondering if this type of idea has implemented before, learning off of only wins and losses.

Data science is a multidisciplinary practice that involves computer programming, advanced mathematics, artificial intelligence, data visualization, database administration, data warehousing, and business intelligence. If you're an expert in only one or two of these subjects, it's a good time to expand your …