Data Science newsletter – November 9, 2016

GROUP CURATION: N/A

In order for the fintech boom to realize its full potential, a number of barriers must be overcome in the way that fintechs relate to investment banks and the CM ecosystem as a whole. These hurdles exist in the areas of simplifying IT architecture, developing industry standards, improving collaboration among players, and mitigating the risks of working with vendors, among others. Moreover, inertia on the part of incumbents can have a dire consequence: the inability to compete with new entrants that use cutting-edge technologies to reverse banks’ traditional competitive advantage. Market structure changes brought on by technology, regulation, and shifting client needs make it critical for incumbent banks to take action now.

A data science round table brought representatives from a dozen industries to the University of Rochester last week to explore opportunities for collaboration.

Rob Clark, the University provost and senior vice president for research, described the round table as a “framing discussion” about how the University’s data science resources “can best serve the Rochester community and raise all our opportunities in the process.”

A group of researchers from New York University and Ohio State University are training the microphones to recognize jackhammers, idling engines and street music, using technology originally developed to identify the flight calls of migrating birds. Think of it as the Shazam, the smartphone app that can identify songs, of urban sounds.

Snippets of audio, about 10 seconds each, will be collected during random intervals over the course of about a year to capture seasonal notes, like air-conditioners and snowplows. The cacophony will be labeled and categorized using a machine-listening engine called UrbanEars. The sensors will eventually be smart enough to identify hundreds of sonic irritants reverberating across the city.

So what is the best possible computer for running a neural network? This is actually an incredibly deep question of physics and computer science. We can ask a related question first: what is the fastest computer that can simulate pouring water into a glass? The answer is an actual glass of actual water, actually being poured. The physical system is a perfect simulator of itself and the computation happens through the physical dynamics. (This is also why no physicist is ever surprised by the statement that the universe is a computer, because it quite literally is.)

The only problem is that it is really hard to measure certain properties of fluids, so people who need this information (aircraft designers, climate scientists, subsea hydraulics engineers) do physical experiments as well as computer simulations.

What does a DataKind project look like before it’s a project? It could be the widened eyes of an “aha” moment during a brainstorm or crossed eyes trying to make sense of a messy dataset. No matter what the path, all projects are born out of conversations between experts that normally don’t get to talk. From Singapore to San Francisco, see how our community events make these conversations possible and incubate projects from a twinkle in the eye to a fully fledged scope of work.

Using machine learning, analytics, and other advanced technologies, Knewton adapts to how students learn, and is able to adjust what content students see based on their strengths and weaknesses. Students can track their progress, work with a study buddy, and receive helpful tips from the teacher.

NK) You have pioneered the research of finding global optimal in nonconvex problems, something that has been a big headache for every machine learning user. You have proved optimality in tensor factorization problems. Can you mention other areas where the community has found algorithms that find global optima?

It turns out that many non-convex machine learning problems can be solved using computationally efficient algorithms. In addition to tensor factorization, this includes matrix completion, robust PCA, phase retrieval, dictionary learning, and so on: the list is growing. In contrast to traditional computer science theory, where the focus is on solving worst-case instances, in machine learning, our goal is solve only a limited class of problem instances. Thus, instead of studying the hardness of solving the worst-case instance, the focus is on characterizing conditions under which finding global optima becomes tractable. For the problems mentioned above, those conditions turn out to be quite natural and mild.

In the digital era, data is the fundamental currency. How organizations handle it throughout the data supply chain—from collection, aggregation, sharing and analysis, to monetization, storage and disposal—can have a decisive impact on their reputation and effectiveness. A data supply chain framework helps practitioners evaluate current ethical practices and implement appropriate ethical controls at each step.

The Moore Foundation asked attendees to describe a current favorite tool or method, and listed the items publicly.