NYU Data Science newsletter – February 16, 2016

GROUP CURATION: N/A

With ICLR, deep learning’s premiere conference, neural network pioneers Yann LeCun and Yoshua Bengio have undertaken a grand experiment in academic publishing. Embracing a radical level of transparency and unprecedented public participation, they’ve created an opportunity not only to find and vet the best papers, but also to gather data about the publication process itself. Last week, conference organizers announced 2016’s accepted papers. In ICLR’s spirit of experimentation, I’ll try to analyze the results, taking a look at what worked and also what didn’t.

… To understand why gravitational waves present a data challenge, we need to know how LIGO works. LIGO is probably the most high-tech and precise instrument human beings have ever built. It is designed to measure minute changes in length, caused by minute stretching or compression of space-time as gravitational wave passes by. The LIGO is so precise it can measure a motion 10,000 times smaller than an atomic nucleus?—?it is the smallest measurement ever attempted by science.
Two LIGO sites in the USA?—?2000 miles from each other to minimize accidental signals

But that level of accuracy presents its own problems. The instrument is prone to noises. A car driving by on highways miles away, scientists walking around, planes flying by, random seismic events, etc. They all produce vibrations much greater than potential gravitational wave signals. A significant part of LIGO engineering challenge is to develop hardware and software to filter out those noises. That is why there are 1000+ co-authors on the first LIGO paper, many of whom are engineers.

In this work we explore recent advances in Recurrent Neural Networks for large scale Language Modeling, a task central to language understanding. We extend current models to deal with two key challenges present in this task: corpora and vocabulary sizes, and complex, long term structure of language. We perform an exhaustive study on techniques such as character Convolutional Neural Networks or Long-Short Term Memory, on the One Billion Word Benchmark. Our best single model significantly improves state-of-the-art perplexity from 51.3 down to 30.0 (whilst reducing the number of parameters by a factor of 20), while an ensemble of models sets a new record by improving perplexity from 41.0 down to 23.7. We also release these models for the NLP and ML community to study and improve upon.

What motivates people to become fastidious custodians of their own health? Clearly, a medical emergency – a cancer diagnosis, heart attack, or trip to the hospital – can be a powerful motivator. However, I left my university job thinking more broadly about the science of behavior change. Can tracking basic biology create positive reinforcements for health-related behaviors?

Twenty years ago, graph paper, a pencil, and commitment were the only option. Today, however, the solution is something everyone has in their pocket, or on their wrists. Modern wearable sensors can track heart rate, calories burned, sleep patterns and so much more. The data and trends can be presented on our computers and our phones every day.

Stephen Pratt, the former InfoSys executive who joined IBM nine months ago to boost its cognitive computing efforts, has left the company, The Wall Street Journal reported Wednesday.

A team of engineers uses computational analysis tools to scrape information from police-related incidents to reveal discrimination and reduce crime.

… who cares for the caretaker? Those with the highest levels of throughput. At Kip, we’re designing a “machine assistant for human assistance.” Instead of trying to replace the human element, we enable this social behavior, streamline it and make it more efficient. AI assistants are then a natural extension of our routing, where machines can curate this endless data overload into manageable chunks to assist our decision making.

Just like how calculators assist us in complex calculations, or computers assist us in knowledge, AI can assist in optimizing connections. AI assistants can help us collect information on a situation more quickly, asses the competing needs of individuals and come up with several scenarios where we choose the best course of action. Armed with more information, we can make better decisions not just for ourselves, but for our communities as well.

Our goal is to invite innovative, fresh concepts, approaches or models for advancing the field. Presentations previously made at the Games for Change Festival or similar forums are unlikely to be seriously considered. The same goes for self-promotional presentations that are not focused on transparent sharing of lessons learned, failures, and meaningful takeaways for the audience.

Deadline: Call for Speakers, Talks, Presentations, and Demos ends on Wednesday, February 24.

While there has been significant progress in [IoT], there remain significant challenges in terms of (1) interoperability and a standardized modular systems architecture, (2) privacy, security and user safety, as well as (3) how users interact with, manage and control an ensemble of devices in this connected environment.

It is in this context that we are happy to invite university researchers1 to participate in the Internet of Things (IoT) Technology Research Award Pilot. This pilot provides selected researchers in-kind gifts of Google IoT related technologies (listed below), with the goal of fostering collaboration with the academic community on small-scale (~4-8 week) experiments, discovering what they can do with our software and devices.

Deadline for proposals is Monday, February 29.

We seek a limited number of student volunteers for the 2016 Conference of the North American Chapter of the Association for Computational Linguistics – Human Language Technology.

Deadline to apply is Wednesday, March 16.

The Intelligence Advanced Research Projects Activity (IARPA) is seeking information on novel training datasets and environments to advance artificial intelligence (AI). This request for information (RFI) is issued solely for information gathering purposes; the RFI does not constitute a formal solicitation for proposals. IARPA anticipates that responses to this RFI will be used to inform future funding opportunities for creating novel training resources for artificial intelligence algorithms. The following sections of this announcement contain details of the scope of technical efforts of interest, along with instructions for the submission of responses.

Deadline for submissions is Friday, April 1.