NYU Data Science newsletter – September 14, 2016

GROUP CURATION: N/A

This was one of those well-written articles that neatly pulls together a lot of points being made around the tech-o-sphere and puts them in one place. Syllabus writers, take note.

Nvidia Corp. announced new processors Monday to try to embed its products in artificial-intelligence systems that are increasingly becoming part of daily life.

The chipmaker, which dominates in video gaming, rolled out graphics chips for running software that makes split-second decisions needed when everything from phones to cars to internet search engines respond to inputs such as speech, images and moving objects.

When Bruce says someone is trying to take down the internet, we should all listen. He knows cyber security and has been seeing, “precisely calibrated attacks designed to determine exactly how well [big] companies can defend themselves, and what would be required to take them down. We don’t know who is doing this.”

It is estimated that about 80 percent of Zika infections are asymptomatic or have symptoms so mild that the disease is not detected. This means the number of cases reported by disease surveillance systems in the U.S. and across the world might be only a small fraction of the actual number of infections. In fact, it’s likely we are are underestimating imported cases in the U.S. and even likely some locally spread cases.

In this situation, mathematical and computational models that account for mosquito populations, human mobility, infrastructure and other factors that influence the spread of Zika are valuable because they can generate estimates of the full extent of the epidemic.

Four years ago, I wrote that when one wants to predict the future, it is always a good thing to rely on the steamrollers, i.e. the exponential trends on which we can surf to predict the future because they are highly efficient. Let see how this trend continues.

‘Reproducibility editor’ Victoria Stodden explains the growing movement to make code and data available to others.

Your next financial adviser might be a centaur — not half-human, half-horse, but half-human, half-machine. … Here, machine learning isn’t about picking stocks, but about predicting human behavior.

Human genome sequencing costs have dropped precipitously over the last few years, however the analytical ability to meet the growing demand for making sense of large data sets remains as a bottleneck. With the introduction of ‘leaner and meaner’ sequencers several years ago, there has been an exponentially increasing need to expand the number of human genomes and data volume for biomedical studies, both academically and commercially. Computational solutions, both software and hardware, are severely needed.

In a study published in BMC Bioinformatics, researchers from Baylor College of Medicine’s Human Genome Sequencing Center, along with Oak Ridge National Laboratory, DNAnexus and the Human Genetics Center at the University of Texas Health Science Center, have developed a novel hybrid computational strategy to address this challenge when processing large data sets.

Microsoft researchers have reached a milestone in the quest for computers to understand speech as well as humans.

Xuedong Huang, the company’s chief speech scientist, reports that in a recent benchmark evaluation against the industry standard Switchboard speech recognition task, Microsoft researchers achieved a word error rate (WER) of 6.3 percent, the lowest in the industry.

More than 40 neuroscientists gathered at the University of Washington eScience Institute for the first-ever Neurohackweek, held Sept. 5 – 9. Inspired by AstroHackWeek and previous brainhack.org events, the week was part conference, part summer school (including tutorials on cloud computing with Amazon Web Services, image processing with open source tools, modeling and statistics, and many others), and in large part focused on group work on novel computational projects, or “hacks” in human neuroscience.

ERIC LEININGER: We’ve seen marketing shift to become more personalized, engaging customers in a different kind of relationship than in the past. What’s surprising to you about the way this marketing is changing?

TOM O’TOOLE: It may not be fully appreciated how advanced data-driven targeted marketing has become.

At United, for example, we are currently doing a targeted marketing program, particularly to grow our market share in what are called “key spoke” cities, meaning important cities that are not our hubs. This is the largest such program we’ve undertaken, targeted to several million individuals. Each individual received, and is now tracking toward, a multi-tiered personalized offer based on her or his actual flights, spending on United, and other factors in the last year.

Scientists have now shown that dolphins alter the volume and frequency of pulsed clicks to form individual “words” which they string together into sentences in much the same way that humans speak.

Also in computational linguistics and speech recognition:

Microsoft researchers achieve speech recognition milestone (September 13 Richard Eckel for the Microsoft blog)

Brain-sensing technology allows typing at 12 words per minute (September 12, Amy Adams for Stanford News)

A nose by any other name: Biology may affect the way we invent words (September 12, Sarah Kaplan for The Washington Post, Speaking of Science blog)

WaveNet: A Generative Model for Raw Audio (September 08, Aaron van den Oord, Heiga Zen, and Sander Dieleman for Google DeepMind blog)

Technology for reading signals directly from the brain developed by Stanford Bio-X scientists could provide a way for people with movement disorders to communicate.

The Cybathlon will introduce “pilots” with spinal-cord injuries and amputations who will control robotic devices to navigate obstacles in six disciplines. Technologies such as powered exoskeletons, powered prostheses (arm and leg), functional electrical-stimulation bikes, powered wheelchairs and brain-computer interfaces will each get their own race track in the middle of the arena.

Instead of the usual goalposts that stand on each end of the rink, the floor will be set up with a sequence of obstacles along the center for the pilots wearing exoskeletons or prosthetic limbs. Another track will be laid out with machines for the brain-computer interface race. And a bigger racetrack will wrap around the rink for the pilots on bikes.

Humans across the globe may be actually speaking the same language after scientists found that the sounds used to make the words of common objects and ideas are strikingly similar.

The discovery challenges the fundamental principles of linguistics, which state that languages grow up independently of each other, with no intrinsic meaning in the noises which form words.

But research which looked into several thousand languages showed that for basic concepts, such as body parts, family relationships or aspects of the natural world, there are common sounds – as if concepts that are important to the human experience somehow trigger universal verbalisations.

Baidu Inc, the Chinese internet search giant, has created a US$200 million venture capital unit to invest in artificial intelligence projects.

The new unit, Baidu Venture, is chaired by chief executive Robin Li Yanhong and will focus on projects also in augment reality and virtual reality.

“The first-phase investment is planned at US$200 million and the fund will invest in projects that are at their early stage,” the company said.

Deep learning papers represented only ~0.15% of computer science papers in arXiv published in early 2012 but grew ~10X to ~1.4% by the end of 2014. In 2016, 80% of papers at many top NLP conferences are deep learning papers. Deep nets are now demonstrating state of the art results across applications in computer vision, speech, NLP, bioinformatics, and a growing list of other domains.

Although deep nets are currently non-interpretable, we’ve seen interesting work in progress to compose deep nets with classical symbolic AI to treat logic programs as a prior structure on the network. This approach may allow deep nets to learn logic programs that ultimately yield new scientific insights in a range of fields.

DARPA is soliciting innovative research proposals in the areas of machine learning and human-computer interaction “to create a suite of new or modified machine learning techniques that produce explainable models that, when combined with effective explanation techniques, enable end users to understand, appropriately trust, and effectively manage the emerging generation of Artificial Intelligence (AI) systems.”

To submit yours, simply use your Google account to upload your file on YouTube. Make sure to add a pithy title and a concise description that includes a link to the data set (or sets) you’ve used for your story. Mark your video “unlisted” and under the “advanced settings” tab, choose the option “allow embedding.” Next, complete the entry form when it becomes available (13 February, 2017).

Andrej Karpathy
Hal Daume III
Jack Clark – Mapping Babel
Andrew Gelman – statistics
Fast ML – Learning ML
FlowingData – dataviz
Renee Teate – tons of data sci resources
Data Machina – datasci newsletter
Data Elixir – datasci newsletter
Talking Machines – podcast
Data Stories – podcast
DataTau
Rbloggers
Hacker News

“bayesAB provides a suite of functions that allow the user to analyze A/B test data in a Bayesian framework.”

“Instead of p-values you get direct probabilities on whether A is better than B (and by how much). Bayesian tests are also immune to ‘peeking’ and are thus valid whenever a test is stopped.”

In this post, I will show you how to create an animation using R and ffmpeg. The idea to do so is pretty simple:

Generate a number of snapshots

Combine them in a video file using ffmpeg

OpenStreetMap 0.3.3 has been released to CRAN.

The most important update is the ability to use custom tile servers.

What about the kinds of applications which need more than the 4GB GPUs Amazon offers (even their newest g2.8xlarge still offers the same 4GB GPUs, be it x4)? The few other cloud providers offering bigger GPU’s (6GB generally) all seem to be either too custom tailored for very specific applications (video edition or biosci), or just completely unusable.

So what is one to do? Simple: get your own GPU rig!

Students actually showed up, so we really do have to teach the course. MCB112 Biological Data Analysis is now in its first week.

The tricksiest bit in the first couple weeks is bringing people up to speed in writing Python, for people who’ve never written code before. We trust in the power of trial and error. We give working example scripts that are related to what the students are asked to do on a problem set. Developing code by mutation, descent with modification, and selection: coding for biologists.