Smartphones, security cameras, and speakers are just a few of the devices that will soon be running more artificial intelligence software to speed up image- and speech-processing tasks. A compression technique known as quantization is smoothing the way by making deep learning models smaller to reduce computation and energy costs. But smaller models, it turns out, make it easier for malicious attackers to trick an AI system into misbehaving — a concern as more complex decision-making is handed off to machines.
In a new study, MIT and IBM researchers show just how vulnerable compressed AI models are to adversarial attack, and they offer a fix: add a mathematical constraint during the quantization process to reduce the odds that an AI will fall prey to a slightly modified image and misclassify what they see.
USA.gov, the official online portal of the U.S federal government, launched an artificial intelligence (AI) powered chatbot named Sam to automate the process of helping people find information on scams and frauds.
“In early February 2019, we launched the bot on three USA.gov webpages. In over a month we’ve had over 4,000 users, and 78% have successfully asked their question and gotten an answer,” says USAGov’s Marietta Jelks.
Civilian ridesharing traffic, like Lyft, benefits every day from tailored data-driven routing methods, but what about emergency vehicles? As part of a long-term effort, new research from CDS Moore-Sloan Data Science Fellow Anastasios Noulas and Birkbeck College, University of London researchers Marcus Poulton, David Weston, and George Roussos aims to improve ambulance response times with increased capabilities for data-driven route prediction and optimization. This project relates to research Noulas contributed to last year.
The Moore-Sloan Data Science Environments (MSDSE) program, established jointly by the Gordon and Betty Moore Foundation and the Alfred P. Sloan Foundation, was launched in 2014 to foster and enhance data-intensive discovery at academic institutions. The program initially funded three data science environments for five years (recently extended for an additional two years): the eScience Institute at the University of Washington, the Berkeley Institute for Data Science (BIDS) at the University of California Berkeley, and the Center for Data Science (CDS) at New York University. As the MSDSE transitions into its extension period, we are excited to report that UW’s eScience Institute has become a permanent Institute with consistent positive impact on the University of Washington campus, and a bright and sustainable future.
Spotify has opened up a new research and development hub in London that the music streaming giant said will create close to 300 new jobs in the capital.
The Swedish company confirmed it would move staff into the Adelphi Building near the Strand on Tuesday.
Spotif, which is listed in New York and has a market value of over $25bn, has grown rapidly from a music streaming upstart into a global firm with more than 4,000 staff. A spokesman said 550 people would be based in the new London office, up from around 260 current roles.
At universities around the country, data science has spread far beyond its traditional homes in the mathematical sciences and computer science departments. Nearly every discipline has data waiting for discovery. At Binghamton University, the demand for training and scholarship has moved so fast that faculty, students and staff have begun grassroots efforts to build skills and share ideas.
The Data Salon is one example. It’s not a traditional departmental seminar; it’s a talk about a specific topic, followed by discussion that welcomes viewpoints from other disciplines, says Xingye Qiao, associate professor of mathematical sciences and an organizer of the Data Salon.
From “Breaking Bad” to “The Walking Dead,” AMC is known for turning unlikely scripts into runaway success stories. AMC Networks president and CEO Josh Sapan told the audience of Variety’s Silicon Valleywood presented by PwC in Menlo Park, Calif., that the secret to these blockbusters is taking creative risk, and not analytics and data science.
“The analytics related to consumption are wildly abundant and available,” Sapan said on Tuesday. AMC is running a number of streaming services these days, and Sapan said he was able to walk to people in his office, look over their shoulder and see what consumers were doing by the second. But all the data in the world doesn’t help make great TV shows, he argued: “You can’t science your way to great stories.”
A generous gift from American Family Insurance today coupled with a planned expansion of its research collaboration will spearhead the growth of UW–Madison’s research in data science, supporting work that could push forward studies in fields as different as artificial intelligence, genetics, drug development, material science and business. The emerging field of data science has its roots in computer science and statistics and is catalyzing unprecedented growth in the global economy.
American Family Insurance is announcing today that it will invest $20 million in UW data science initiatives, including $10 million in research over the next 10 years — through 2029 — and establish a $10 million endowment to create the American Family Insurance Data Science Institute on campus.
Artificial-intelligence (AI) researchers across all academic career stages are feeling the lure of industry, thanks to higher salaries and perks such as access to large data sets and computing resources. Businesses are recruiting AI specialists for projects ranging from modelling risk in finance to designing crop-harvesting robots.
In the past year, the number of PhD graduates on LinkedIn who say they have AI expertise has risen by 66%, according to a report published in April by the software provider Element AI in Montreal, Canada. And the number of researchers publishing at top machine-learning conferences increased by 19% (see go.nature.com/22nexux). From a sample of 4,500 researchers, the report authors found that the United States continues to be a major hub of AI training and employment. Other hotspots include China, the United Kingdom and Germany. But demand is outstripping supply, says Yoan Mantha, market-intelligence lead at Element AI. The company estimates that in the United States, there are around 144,000 AI-related job openings and only about 26,000 developers and specialists seeking work.
Microsoft scientists and product developers have pioneered a complementary approach called machine teaching. This relies on people’s expertise to break a problem into easier tasks and give machine learning models important clues about how to find a solution faster. It’s like teaching a child to hit a home run by first putting the ball on the tee, then tossing an underhand pitch and eventually moving on to fastballs.
“This feels very natural and intuitive when we talk about this in human terms but when we switch to machine learning, everybody’s mindset, whether they realize it or not, is ‘let’s just throw fastballs at the system,’” said Mark Hammond, Microsoft general manager for Business AI. “Machine teaching is a set of tools that helps you stop doing that.”
An ecologist built an army of beach surveyors over 20 years and now has the world’s largest data set of marine bird mortality informing climate change and disaster studies.
Less than a year remains before the United States undertakes a decennial census for the 24th time in its history. On April 1, 2019, national experts gathered at Brookings for a joint event with the National League of Cities (NLC) Institute to discuss what the 2020 census means not only for the nation as a whole, but also for the major cities and metropolitan areas on the front lines of America’s demographic change. Presenters and panelists highlighted four key takeaways for people and organizations committed to ensuring a full, fair, and accurate count of the U.S. population as preparations continue to ramp up for next year.
There’s a lot at stake in the 2020 census. The panelists noted that the decennial census provides the statistical basis for the distribution of hundreds of federal financial assistance programs. According to
forthcoming research from Andrew Reamer at George Washington University, the federal government provided an estimated $883 billion in funding from 55 large expenditure programs to states in fiscal year 2016 that was based on data derived from the 2010 census.
A powerful New Mexico regulatory authority is requiring the state’s largest utility to bill Facebook $39 million for a new transmission line construction.
The Albuquerque Journal reported that the Public Regulation Commission ordered Public Service Company of New Mexico to charge Facebook for nearly half the cost of the $85 million transmission project for its New Mexico data center that opened this year.
In exponential curves, Singularitarians see super-intelligence and abundance. Most people outside the Singularity bubble believe that natural systems behave like S-curves, where systems respond and self-regulate. When a pandemic has run its course, for example, its spread slows and the world settles into a new equilibrium. The world may not be in the same state as before the pandemic or other runaway change, but the notion of singularity – especially as some sort of saviour or judgment day that will allow us to transcend the messy, mortal suffering of our human existence – is fundamentally a flawed one.
The first biometric airport terminal is already up and running, thanks to Delta’s collaboration with U.S. Customs and Border Protection, the Transportation Security Administration, and Hartsfield-Jackson Atlanta International Airport. Similar systems are now operational at Dubai International Airport, at least for first- and business-class passengers. While groundbreaking, these systems—powered by artificial intelligence—are early exemplars of the emerging technological transformation of airport and airline operations. They are working preludes to artificial intelligence’s eventual forays throughout the passenger experience. Part of these forays will be driven by technology; we’ll have increasingly smarter and capable tools for independently managing enormous complexities and creating new efficiencies. Another part will be driven by new competitive pressures; in the face of autonomous vehicles that will offer compelling alternatives to flying, the future of air travel will require robust improvements.
Taken together, we can reasonably anticipate artificial intelligence to finally solve for some of the most intractable problems that have frustrated passengers and vexed airlines and airports for decades.
University of California-Berkeley, Berkeley Institute for Data Science
from
Berkeley, CA May 1, starting at 3:10 p.m., University of California-Berkeley (190 Doe Library). Speaker: Shreeharsh Kelkar from University of California-Berkeley. [free]
I’ve been very busy working on my MRes project in recent weeks, having very little sleep. This made me seek ways to improve my workflow in the most important tool of my work: Jupyter Notebook/Jupyter Lab. I collected all the hacks & tips in this piece, hoping that other researchers may find those useful.
Too often however, data science technical debt is more like a payday loan. We take shortcuts in developing a solution without an understanding of the risks and costs of those shortcuts, and without a realistic plan for how we’re going to pay back the debt. Code is produced, but it’s not tested, documented, or robust to changes in the system. The result is that data science projects become expensive or impossible to maintain as time goes on.
In April 2019 I decided to annotate my funded NSF CAREER proposal with a bunch of margin notes. I’m making this proposal public because I think it’s the most equitable when everyone has access to good examples. I hope you find it useful, but if I don’t know you please don’t email me to ask for clarifications, materials, or feedback on your proposal.