I'd love to talk more about Flowcast, but I'm still not able to shake the image of you making a robotic submarine run by San Diego poolside (laughs). As a STEM enthusiast, I have been in awe of IBM Watson's capabilities. And I feel it's an honor to be talking to someone who has contributed to its capabilities. Now, let's come back to Flowcast. Can you share more information and shed more light on how Flowcast came about?

Microsoft executive Todd Holmdahl has led teams to invent profitable new computing hardware products before. His latest project is his first with a chance of hauling in a Nobel Prize in physics as well as new revenue if it succeeds. Holmdahl previously oversaw the hardware design of the Xbox and Xbox360 consoles, which rake in billions for Microsoft each year. Late last year he was appointed the leader of a swelling band of mathematicians, physicists, and engineers trying to add mighty computers powered by quantum physics to Microsoft's menu of cloud computing services. Holmdahl speaks about quantum computing like a tech executive would a new line of business, not a speculative physics or R&D project.

By Andrew Trask This tutorial teaches backpropagation via a very simple toy example, a short python implementation. Consider trying to predict the output column given the three input columns. We could solve this problem by simply measuring statistics between the input values and the output values. If we did so, we would see that the leftmost input column is perfectly correlated with the output. Let's jump right in and use it to do this.

In practice, given a wide range of classifiers, we often have to choose the one based on performance comparison through validation. Research literature shows that there is no classifier that performs universally best in all contexts for all problems. The following paper applied 8 most popular classifiers (e.g., SVM, Neural Net, Ensemble, KNN, Decision Tree, Logistic Regression, Discriminant Analysis, Naive Bayes, etc.) in Machine Learning arena to solve a problem currently confronting finance institutions such as banks, insurers, asset managers in their derivative valuation and risk management. The paper shows that when properly parameterized (which the paper discusses in details), the performances are either consistent with or contrary to some classic studies in the area. The paper is available at SSRN.

Most innovation strategies today encompass emerging technologies, including artificial intelligence (AI). As the business advantage of the future, according to 72% of business decision makers recently surveyed by PwC, artificial intelligence and its value are critical concepts for business leaders to grasp. AI is technology that can understand, learn, and then act based on gained or gathered information, holding the potential to transform markets far and wide. Businesses are betting big: 54% of executives in our 2017 Global Digital IQ Survey tell us their companies are making substantial investments in AI today; in three years, that number increases to 63%. The rise of AI comes at a point when many companies already are feeling pressure to reinvent.

The algorithm learns patterns through examining combinations of factors that lead to suicide, from medication use to the number of ER visits over many years. In a trial run in Montana, which often ranks near the top among states with the highest suicide rate, the team found 5 percent of people visiting the ER were at high risk. "Proxi is based on the idea that mental health happens outside the direct connection with a clinician, within friends and family who are an individual's natural support system," Hudenko says. If a patient is classified as high risk, a nurse will help them get set up with the Proxi app to create a support system of friends and family, Hudenko adds.

Female vampire bats are'consoled' by unrelated friends when a family member goes missing, new research has revealed. Despite their malevolent reputation, female vampire bats form strong social bonds with their mothers and daughters as they groom and share regurgitated meals of blood. They also form friendships with less closely related bats and researchers discovered that unrelated friends are an important source of support when family members go missing. Researchers have found that when major food donors, like mothers or daughters, were removed from a bat's social network, females who previously built up more friendships with non-relatives coped better with their loss. Researchers from the Smithsonian Tropical Research Institute in Panama City monitored social interactions in a captive colony of around 30 marked common vampire bats for four years.

We introduce dropout compaction, a novel method for training feed-forward neural networks which realizes the performance gains of training a large model with dropout regularization, yet extracts a compact neural network for run-time efficiency. In the proposed method, we introduce a sparsity-inducing prior on the per unit dropout retention probability so that the optimizer can effectively prune hidden units during training. By changing the prior hyperparameters, we can control the size of the resulting network. We performed a systematic comparison of dropout compaction and competing methods on several real-world speech recognition tasks and found that dropout compaction achieved comparable accuracy with fewer than 50% of the hidden units, translating to a 2.5x speedup in run-time.

Event sequence, asynchronously generated with random timestamp, is ubiquitous among applications. The precise and arbitrary timestamp can carry important clues about the underlying dynamics, and has lent the event data fundamentally different from the time-series whereby series is indexed with fixed and equal time interval. One expressive mathematical tool for modeling event is point process. The intensity functions of many point processes involve two components: the background and the effect by the history. Due to its inherent spontaneousness, the background can be treated as a time series while the other need to handle the history events. In this paper, we model the background by a Recurrent Neural Network (RNN) with its units aligned with time series indexes while the history effect is modeled by another RNN whose units are aligned with asynchronous events to capture the long-range dynamics. The whole model with event type and timestamp prediction output layers can be trained end-to-end. Our approach takes an RNN perspective to point process, and models its background and history effect. For utility, our method allows a black-box treatment for modeling the intensity which is often a pre-defined parametric form in point processes. Meanwhile end-to-end training opens the venue for reusing existing rich techniques in deep network for point process modeling. We apply our model to the predictive maintenance problem using a log dataset by more than 1000 ATMs from a global bank headquartered in North America.

Contextual bandits are a form of multi-armed bandit in which the agent has access to predictive side information (known as the context) for each arm at each time step, and have been used to model personalized news recommendation, ad placement, and other applications. In this work, we propose a multi-task learning framework for contextual bandit problems. Like multi-task learning in the batch setting, the goal is to leverage similarities in contexts for different arms so as to improve the agent's ability to predict rewards from contexts. We propose an upper confidence bound-based multi-task learning algorithm for contextual bandits, establish a corresponding regret bound, and interpret this bound to quantify the advantages of learning in the presence of high task (arm) similarity. We also describe an effective scheme for estimating task similarity from data, and demonstrate our algorithm's performance on several data sets.