NYU Data Science newsletter – January 12, 2016

GROUP CURATION: N/A

Here’s the quietest story at CES: suddenly everything can move itself around without any humans involved.

Self-driving cars might be the flashy future story that lands on the local news, but look a tick smaller and hey — the dream is already here. The drone revolution was already about aircraft that can stabilize themselves with no user input, but this year the best new drones have computer vision systems that can sense their environment and navigate a path while avoiding obstacles autonomously. Intel is showing off robots with RealSense cameras that can wander around your house and recognize different people. And yes, the cars are getting ever and ever closer to eliminating the steering wheel and basically becoming gigantic rolling movie theaters.

In the talk below, Recursive Deep Learning for Modeling Compositional and Grounded Meaning, Richard Socher, Founder, MetaMind describes deep learning algorithms that learn representations for language that are useful for solving a variety of complex language tasks. He focuses on 3 projects: (i) Contextual sentiment analysis (e.g. having an algorithm that actually learns what’s positive in this sentence: “The Android phone is better than the iPhone”); (ii) Question answering to win trivia competitions (like IBM Watson’s Jeopardy system but with one neural network); (iii) Multimodal sentence-image embeddings to find images that visualize sentences and vice versa (with a fun demo!). All three tasks are solved with a similar type of recursive neural network algorithm. [video, 15:54]

Some time ago, I tried to scrape every Bay Area profile off LinkedIn until the site blocked my entire office network (Lesson learned: Use a proxy). This was Bad because we were (and still are!) hiring.

The goal was to collect enough data to create a set of classifiers that could estimate a person’s salary from their LinkedIn profile.

… a team of astronomers told the American Astronomical Society meeting here that they’ve found a cloud of gas that has tiny amounts of heavy elements, just as you would expect if primordial stars—so-called population III stars—had burned out, exploded, and spread their ingredients through the previously pristine gas cloud. To probe the cloud, the team used an even more distant quasar—a hugely bright light source powered by a supermassive black hole—as a backlight.

Auerbach plans to apply computational chemistry, using computers to reveal the microscopic structures and motions of molecules, to understand how carbohydrates react in zeolite pores. “Computational chemistry provides the most powerful microscope known to humanity, revealing the atomic dance of making fuels,” he says.

The OpenAI research team will be answering your questions.

We are (our usernames are): Andrej Karpathy (badmephisto), Durk Kingma (dpkingma), Greg Brockman (thegdb), Ilya Sutskever (IlyaSutskever), John Schulman (johnschulman), Vicki Cheung (vicki-openai), Wojciech Zaremba (wojzaremba).

… As some investors press Yahoo to fire her, Ms. Mayer is crafting a last-ditch plan to streamline the company — including significant layoffs — that is expected to be announced before month’s end.

While many Yahoo workers are keeping their heads down, just doing their jobs, others have lost faith in Ms. Mayer’s leadership, according to conversations with more than 15 current and former employees from all levels of the company, most of whom spoke on the condition of anonymity because of continuing ties to Yahoo and its strict policy against leaks.

The next big thing in artificial intelligence could be right in Boston’s backyard, and it might just be better than Apple’s Siri and Google Now. Actually, if things go as planned, it’ll blow Siri and the like out of the water.

Semantic Machines, a Newton-based startup with half of its team in Berkeley, Calif., may not have much to show for its conversational AI platform yet besides a few use case examples posted to its website. But what makes the startup seem so promising is its 23-member team, which includes Larry Gillick, Apple’s former chief speech scientist for Siri, along with people who worked on Google Now.

Why does misinformation spread so quickly on the social media? Why doesn’t it get corrected? When the truth is so easy to find, why do people accept falsehoods?

A new study focusing on Facebook users provides strong evidence that the explanation is confirmation bias: people’s tendency to seek out information that confirms their beliefs, and to ignore contrary information.

Americans work some of the longest hours in the Western world, and many struggle to achieve a healthy balance between work and life. As a result, there is an understandable tendency to assume that the problem we face is one of quantity: We simply do not have enough free time. “If I could just get a few more hours off work each week,” you might think, “I would be happier.”

This may be true. But the situation, I believe, is more complicated than that. As I discovered in a study that I published with my colleague Chaeyoon Lim in the journal Sociological Science, it’s not just that we have a shortage of free time; it’s also that our free time, in order to be satisfying, often must align with that of our friends and loved ones. We face a problem, in other words, of coordination. Work-life balance is not something that you can solve on your own.

When assessing a patient, medics look at snapshots of physiological data that are manually taken by doctors or nurses, and make decisions against patient history, family background and test results, as well as their own knowledge and experience. But what if this data was constantly being taken, every second of every day? And what if a system was clever enough to compare these readings to thousands of patients worldwide with a similar history and disorder, as well as all the current clinical guidelines and studies, and make clinical suggestions to doctors?

In 2016, this kind of data-led decision-making will come ever closer. Sentrian, a California-based early-stage machine learning and biosensor analytics company for remote patient management, has created a system that does just that, and it’s currently being trialled on patients.

The 1st Workshop on Representation Learning for NLP (RepL4NLP) will be held on August 11, 2016 and hosted by the 54rd Annual Meeting of the Association for Computational Linguistics (ACL) in Berlin, Germany. The workshop is being organised by Phil Blunsom, Kyunghyun Cho, Shay Cohen, Edward Grefenstette, Karl Moritz Hermann, Laura Rimell, Jason Weston, and Scott Wen-tau Yih, and will be sponsored by Google DeepMind, Facebook AI Research, and Microsoft Research.

Berlin, Germany. Deadline for paper submission: Sunday, May 8

We are excited to announce the release of the latest minor revision of the Jupyter Notebook, version 4.1.

As usual, you can update using pip: pip install notebook –upgrade

… Today there are a growing number of online services that provide the tools such an AI would rely upon. To name a few: voice recognition is available through Amazon’s Alexa Voice Service, facial and emotion recognition via Microsoft’s Project Oxford, natural language processing for understanding what is said and written via IBM Watson Services, and machine learning via the Amazon Web Services and Microsoft Azure cloud platforms.

Not only are these services available to anyone but many are initially free for personal use, for instance Microsoft’s Project Oxford allows developers to make 5,000 free calls a month to its computer vision API.

You also don’t need be a programming genius to piece together a simple app drawing on these services.