A startup with ties to Amazon is sampling a 16-nm chip mainly targeted for data centers that it claims handily beats CPUs and GPUs for deep-learning inference jobs. Habana is raising funds to support its production and a roadmap that includes a 16-nm training chip sampling next year as well as follow-on 7-nm products.
The startup is the latest to join a frothy AI sector of as many as 50 companies with some form of machine-learning accelerator. To date, big data centers driving the technology typically run their workloads on the large banks of CPUs and GPUs that they maintain.
As bike traffic surges in cities across North America, the call for increased safety gets louder every year.
Most of the discussion centres on better infrastructure, including more dedicated bike paths, yet a large part of injury prevention is learning more about where and why cycling accidents happen.
The city of Boston did just that, inviting a team from the Harvard Injury Control Research Center to dive into bicycle accident data and create a map identifying areas in the city where accident rates are the highest.
The job was a big one and required a large number of volunteer university students with experience in statistics to quantify, code and interpret the data in police reports, including written statements by officers and ambulance technicians who were present at a crash. The result was a picture of how safe it is to cycle the streets of Boston, based on four years (2009-2012) of data collected from 1,797 bike crashes.
Big data, machine learning, and interoperability are all topics we’ve been hearing about for many years in health tech. But in fact these banner ideas are deeply intertwined with one another. Machine learning requires a certain amount and type of big data — and interoperability plays a large role in whether or not researchers can gain access to the data they need to train models effectively.
At a panel discussion at Health 2.0’s Provider Symposium, moderated by Healthbox Chief Medical Officer Eric Louie, different speakers weighed in on this complex web of issues.
“We’ve got huge oceans of data both in our electronic medical record systems and outside of the system, but we’re still able to derive just little bubbles of clinical meaning from that data,” Matt Menning, director of IHMI engagement at the American Medical Association, said. “Our goal is not just to start to agree on terms and structure for exchanging data, but also get to some standards around what data we’re collecting.”
Bloomberg BusinessWeek, Hyperdrive; John Lippert, Bryan Gruley, Kae Inoue, and Gabrielle Coppola
from
John Leonard strolls up to a drab one-story garage on the campus of MIT and unlocks the door. The building is remarkable only for the reflective windows, which make it impossible to peek inside. “If you were driving past in a taxi,” Leonard says, “would you think the future of Toyota is being designed here?”
Inside squats a silver Lexus LS 600hL sedan. It’s not just any Lexus—this one is vital to Toyota’s effort to develop driverless vehicles. Leonard, vice president for automated driving research at the almost three-year-old Toyota Research Institute, explains how the Lexus is jury-rigged with radars, video cameras, and lasers that can detect, identify, and react to objects up to 200 meters (656 feet) away—twice as far as a year ago. His charges steer the Toyota-built car around Cambridge, Mass., to capture data that can be used to create digital maps and try to extrapolate how a vehicle might behave without a human at the wheel. “It’s sort of like a science project,” Leonard says. “Stay on the road, don’t hit things, don’t get hit.”
If only it were that simple. Toyota Motor Corp., the world’s most valuable automaker, with a market capitalization of $200 billion, is behind in the race to create the vehicles of a maybe-not-so-distant future. Just four years ago, Akio Toyoda, the company’s president, was saying his company would pursue self-driving vehicles only after one beat a human driver—for instance, him—in a marathon road race. He’s not saying that anymore, because Toyota has too much to lose.
Over the past decade, I’ve watched firsthand as the field of artificial intelligence has transitioned from one phase to the other. The 1980s and 1990s were a period of discovery in A.I., one that I participated in through my research on speech recognition at Carnegie Mellon University and Apple. More recently, I participated in the implementation phase through my work as the head of Google China and as an early investor in the Chinese mobile internet.
This movement from discovery to implementation marks a significant shift in A.I.’s center of gravity — away from the United States and toward China. The age of discovery relied heavily on innovation coming out of the United States, which excels at visionary research and moonshot projects. The country’s freewheeling intellectual environment, unparalleled network of research universities and traditional openness to immigrants (such as myself) have for decades made it an incubator for big ideas in A.I.
A.I. implementation, however, plays to a different set of strengths, many of which are manifested in China: abundant data, a hypercompetitive business landscape and a government that actively adapts public infrastructure with A.I. in mind. China also excels at turning an abstract scientific breakthrough into thousands of useful and commercially viable products.
Despite being a slow-moving disaster, famine is notoriously difficult to predict.
The reason for this, experts say, is that severe food shortages are hardly ever about food supply alone.
A famine might be triggered by drought or some other climatic interference in crop production, but other powerful forces usually bring the scourge to full bloom: food price inflation, political instability, military conflict and even too much rain.
“The root cause of famine is extremely complex,” said Franck Bousquet, senior director of the World Bank Fragility, Conflict, and Violence Group (FCV). “Usually, the poorest and most vulnerable are the most affected and the least able to cope with shocks that other populations can absorb. Out of the last 10 major famines, nine have resulted from conflict and war.”
Harvard University raised a record $9.6 billion as the world’s richest school completed a campaign that became the signature accomplishment of Drew Faust, the president who stepped down this year.
The haul eclipsed a goal of $6.5 billion Harvard set when it kicked off the campaign five years ago. The total set a new high water mark in higher education, which has seen fundraising increase with the proliferation of mega-donations. The gifts to Harvard included $400 million from hedge fund manager John Paulson and $350 million from real estate developer Gerald Chan and his family.
U.S. colleges led by Harvard and Stanford University reaped a record $43.6 billion in charitable contributions in fiscal 2017, according to the Council for Aid to Education, which tracks university giving. Less than 1 percent of all colleges accounted for almost 30 percent of the total raised in the year through June 30, 2017, the latest figures available.
The Washington Post, Eli Rosenberg and Herman Wong
from
A Cornell professor whose buzzy and accessible food studies made him a media darling has submitted his resignation, the school said Thursday, a dramatic fall for a scholar whose work increasingly came under question in recent years.
The university said in a statement that a year-long review found that Brian Wansink “committed academic misconduct in his research and scholarship, including misreporting of research data, problematic statistical techniques, failure to properly document and preserve research results, and inappropriate authorship.”
Wansink, a marketing professor at Cornell’s business college who was the director of the university’s Food and Brand Lab, will retire at the end of the academic year, the school said.
A new government-funded research project at Worcester Polytechnic Institute aims to use artificial intelligence in hopes of revolutionizing the way teachers evaluate their own performance in the classroom.
Computer science professor Jacob Whitehill and his colleagues have received a $750,000, three-year grant from the National Science Foundation to develop the Automatic Classroom Observation Recognition Neural Network platform, or ACORN for short, which will combine machine learning, natural language processing, and elements of psychology and educational theory to deliver rapid feedback on teacher-student interactions.
“I’ve always loved teaching – I really enjoy the dynamics between myself and my students,” Mr. Whitehill said, adding that he’s been particularly fascinated with finding a way to identify the characteristics of a positive exchange between instructor and pupil.
Today’s artificial intelligence systems, including the artificial neural networks broadly inspired by the neurons and connections of the nervous system, perform wonderfully at tasks with known constraints. They also tend to require a lot of computational power and vast quantities of training data. That all serves to make them great at playing chess or Go, at detecting if there’s a car in an image, at differentiating between depictions of cats and dogs. “But they are rather pathetic at composing music or writing short stories,” said Konrad Kording, a computational neuroscientist at the University of Pennsylvania. “They have great trouble reasoning meaningfully in the world.”
To overcome those limitations, some research groups are turning back to the brain for fresh ideas. But a handful of them are choosing what may at first seem like an unlikely starting point: the sense of smell, or olfaction. Scientists trying to gain a better understanding of how organisms process chemical information have uncovered coding strategies that seem especially relevant to problems in AI. Moreover, olfactory circuits bear striking similarities to more complex brain regions that have been of interest in the quest to build better machines.
The New York Times, Julie Creswell, Kevin Draper and Rachel Abrams
from
It started with a simple battery.
Around 2009, Amazon quietly entered the private label business by offering a handful of items under a new brand called AmazonBasics. Early offerings were the kinds of unglamorous products that consumers typically bought at their local hardware store: power cords and cables for electronics and, in particular, batteries — with prices roughly 30 percent lower than that of national brands like Energizer and Duracell.
The results were stunning. In just a few years, AmazonBasics had grabbed nearly a third of the online market for batteries, outselling both Energizer and Duracell on its site.
Inside Amazon’s Seattle headquarters, that success raised a tantalizing possibility. If, with very little effort, Amazon could become a huge player in the battery market, what else might be possible for the company?
Data science is a burgeoning field. As a result of recent technological advances, widespread and accelerated uptake of data-science technologies by many sectors, and the increasing workforce demands for data scientists, a growing number of universities and colleges in the US and abroad are creating academic data-science programs. In order to take stock of these educational initiatives, the Data Science Institute convened the first ever Data Science Leadership Summit in March 2018. Sixty-five leaders in data-science education from 29 public and private universities and three funding organizations gathered at Columbia University in order to accomplish the following:
To initiate the formation of an academic community for data science;
To share best practices among academic leaders who face similar challenges and opportunities; and
To take collective responsibility in the broader effort to prepare next-generation data scientists to contribute in the best interests of society.
China’s economic data, never easy to track in the best of times, have become almost indecipherable in 2018. While the headline numbers on GDP growth, retail sales and debt tell reassuring stories that (unsurprisingly) match government objectives, the underlying statistics paint a very different and conflicting picture. What does this tell us about the state of the economy?
Market watchers constantly pore over a range of numbers, looking for irregularities. From short sellers of Chinese companies listed overseas to academic economists developing creative methods to mirror GDP growth, there’s no dearth of analysts and no lack of data to peruse.
Death rates from drug overdoses in the U.S. have been on an exponential growth curve that began at least 15 years before the mid-1990s surge in opioid prescribing, suggesting that overdose death rates may continue along this same historical growth trajectory for years to come, according to a University of Pittsburgh Graduate School of Public Health analysis published today in Science.
The type of drug and the demographics of those who die from overdoses has fluctuated over the years. When the use of one drug waned, a new drug filled in, attracting new populations from different geographic regions at faster rates. These findings suggest that, to be successful, prevention efforts must extend beyond control of specific drugs to address deeper factors driving the epidemic.
“The current epidemic of overdose deaths due to prescription opioids, heroin and fentanyl appears to be the most recent manifestation of a more fundamental, longer-term process,” said senior author Donald S. Burke, M.D., Pitt Public Health dean and UPMC-Jonas Salk Chair of Global Health.
tl;dr: We observed that Dynamic Programming is the common base of both database query optimization and reinforcement learning. Based on this, we designed a deep reinforcement learning algorithm for database query optimization we call DQ. We show that DQ is highly effective and more generally adaptable than any of the prior approaches in the database literature. We feel this is a particularly good example of AI and Database research coming together: both because of the shared algorithmic kernel, and because of the pragmatic need to resort to effective data-driven heuristics in practice. A preprint of the paper–Learning to Optimize Join Queries With Deep Reinforcement Learning–is on arXiv; a more technical blog post is available as well.
We have just announced that a foundation will be supporting scikit-learn at Inria [1]: scikit-learn.fondation-inria.fr
This is an exciting turn for us, because it enables us to receive private funding. As a result, we will be able to have secure employment for some existing core contributors, and to hire more people on the team. The goal is to help sustaining quality (more frequent releases?) and to tackle some ambitious features.