“You can think of curiosity as a kind of reward which the agent generates internally on its own, so that it can go explore more about its world,” Agrawal said. This internally generated reward signal is known in cognitive psychology as “intrinsic motivation.” The feeling you may have vicariously experienced while reading the game-play description above — an urge to reveal more of whatever’s waiting just out of sight, or just beyond your reach, just to see what happens — that’s intrinsic motivation.
Humans also respond to extrinsic motivations, which originate in the environment. Examples of these include everything from the salary you receive at work to a demand delivered at gunpoint. Computer scientists apply a similar approach called reinforcement learning to train their algorithms: The software gets “points” when it performs a desired task, while penalties follow unwanted behavior.
But this carrot-and-stick approach to machine learning has its limits, and artificial intelligence researchers are starting to view intrinsic motivation as an important component of software agents that can learn efficiently and flexibly — that is, less like brittle machines and more like humans and animals.
Apple says its version of the technology, called Face ID and available when the phone ships in November, uses a suite of sensors to map your face in 3-D. An infrared light illuminates your face, and a projector projects an array of infrared dots at it. An IR camera snaps an image of these dots, which the phone uses to authenticate you against an already-stored image of your face. The company claims its Face ID feature is so secure that there’s a one in a million chance that someone could spoof you—not only does it require a measure of the user’s attention to unlock the phone, but Apple also says it trained the feature on realistic-looking 3-D masks so it would not be tricked by them.
The technology behind Face ID is not novel. Anil Jain, a Michigan State University professor who studies biometric recognition and computer vision, notes that it uses an existing tactic called structured light to capture your visage in three dimensions—something he employed for object recognition back in the 1980s.
A publishing trade group has suggested an arrangement with ResearchGate, a scientific networking site, that would restrict unlicensed sharing of copyrighted scientific journal articles.
Founded in 2008, ResearchGate enables scientists to share and discuss papers and to connect with collaborators. It now has more than 13 million members and 100 million publications, according to its website. Its funding comes from venture capital investment; investors include Bill Gates, Goldman Sachs Investment Partners, and Wellcome Trust.
On June 1 2017, President Trump announced his intention to pull the U.S. out of the Paris Climate Change Agreement. His announcement runs contrary to overwhelming domestic support, among voters in every state in the country, for the U.S. government to remain part of the Agreement. His administration’s stance and reiteration of their intent to withdraw has sparked a flurry of support from states, cities, companies, investors, and universities that have signaled their renewed commitment to tackling climate change and redoubled efforts to meet the Paris Agreement’s global climate goals.
Data-Driven Yale (DDY) has mapped the landscape of these climate actions, showing the more than 500 city and state (subnational) governments, representing nearly half the U.S. population, that have committed to climate action. These subnational jurisdictions are joined by nearly 3,000 U.S. businesses with over $7.7 trillion in revenue and over 700 universities (non-state actors) with a total student population nearing 1 million and a collective endowment of over $250 billion.
A group of prominent scientists on Monday created a potential whiplash moment for climate policy, suggesting that humanity could have considerably more time than previously thought to avoid a “dangerous” level of global warming.
The upward revision to the planet’s influential “carbon budget” was published by a number of researchers who have been deeply involved in studying the concept, making it all the more unexpected. But other outside researchers raised questions about the work, leaving it unclear whether the new analysis — which, if correct, would have very large implications — will stick.
In a study published in the journal Nature Geoscience, a team of 10 researchers, led by Richard Millar of the University of Oxford, recalculated the carbon budget for limiting the Earth’s warming to 1.5 degrees Celsius (2.7 degrees Fahrenheit) above temperatures seen in the late 19th century. It had been widely assumed that this stringent target would prove unachievable — but the new study would appear to give us much more time to get our act together if we want to stay below it.
Now, they may be able to something for you — predict your risk of stroke or heart failure.
Scientists at the University of Rochester Medical Center and the Rochester Institute of Technology teamed up to use the camera on your smart device to pick up changes in your skin color that are too subtle for the eye to see.
“I call it a useful selfie,” said Jean-Phillippe Couderc, associate professor of cardiology at URMC, who helped develop and now is testing an app to detect the irregular heartbeat of atrial fibrillation, AF. “Everybody takes a selfie and shows their face. … This is a new way of using this kind of behavior.”
Extra Extra
Robert Lustig is a pediatric endocrinologist with a new book out about “succumbing to pleasure”, increasing rates of addiction, and increasing rates of depression (see this week’s Data Viz of the Week). The FT Alphacat podcast interview with him had me 90% captivated, 10% skeptical.
And Topos asks if you could perchance bypass the MBA by using machine learning to predict the best site locations for opening a new business. They use New York’s coffee shop scene as a case study.
EurekAlert! Science News, University of Texas Arlington
from
Like the organ itself, data collected about the brain is incredibly complex and requires sophisticated methods to sort and analyze. Current methods involve topology, which is a mathematical model that gives a picture of the data, or machine learning, which is a statistical model showing trends. Neither can handle the amounts of data available.
Junzhou Huang, an associate professor in the Computer Science and Engineering Department at The University of Texas at Arlington, will use a $210,000 National Science Foundation grant to explore how to combine the two methods to more accurately predict the outcome of future data. Chao Chen at the City University of New York is co-principal investigator on the project.
The Brookings Institution, Amy Liu and Allison Plyer
from
For public officials to effectively steer a recovery process and for citizens to trust in the effort, reliable, transparent information will be essential. Leaders and the public need a shared understanding of the scale and extent of the damage and which households, businesses and neighborhoods have been affected. This is not a one-time effort. Data must be collected and issued regularly over months and years to match the duration of the rebuilding effort.
Without this information, it will be nearly impossible to estimate the nature of aid required, determine how best to deploy resources, prioritize spending and monitor progress. Rebuilding processes are chaotic, with emotions high over multiple, competing priorities. Credible public information organized in one place can help to neutralize misconceptions, put every need in context and depoliticize decision-making. Most importantly, data on recovery needs also can enable citizen involvement and allow residents to hold public leaders accountable for progress.
In this post I will give my personal thoughts on some articles from 2017 ACM Conference on Recommender Systems, that I chose out of my own interest, with no specific order, divided by recent trends in the area.
In this article we look at New York City through the lens of coffee in an attempt to explore a fundamental question of spatial economics: how are the locations of businesses determined?
The push for open government data got a boost this week with passage of a budget bill that includes language codifying open data requirements for the federal government.
Senate passage of the Defense Department spending authorization package included an amendment incorporating the text of the Open, Permanent, Electronic, and Necessary (OPEN) Government Data Act. If, as expected, the DoD budget bill becomes law, the open data amendment provides a legal mandate for the government to adopt open data practices. How that requirement would be met remains unclear.
Last month, Facebook and Google came out forcefully against a bill that would hold companies accountable for hosting sex trafficking on their websites. They said that while they worked hard to combat sex trafficking, changing the law “jeopardizes bedrock principles of a free and open internet” that have been crucial to innovation for decades.
By this week, the two companies were hoping to reach a compromise with lawmakers, an acknowledgment that they could not stop the bill entirely because of strong political headwinds.
The shifting position illustrates the changing political reality in Washington for some of the country’s biggest technology companies. After years of largely avoiding regulation, businesses like Facebook, Google and Amazon are a focus of lawmakers, some of whom are criticizing the expanding power of big tech companies and their role in the 2016 election.
Now the real work begins for The Engine, MIT’s ambitious venture fund and incubator.
The organization announced its first batch of seven investments on Tuesday (see below), and revealed that it has raised $200 million, with plans to back 40 to 50 so-called “tough-tech” companies over the next few years. The Engine initially raised $150 million for its first fund, but later tacked on the additional $50 million. MIT is one of the investors in the fund; it chipped in $25 million.
Daniel Rothman, professor of geophysics in the MIT Department of Earth, Atmospheric and Planetary Sciences and co-director of MIT’s Lorenz Center, has analyzed significant changes in the carbon cycle over the last 540 million years, including the five mass extinction events. He has identified “thresholds of catastrophe” in the carbon cycle that, if exceeded, would lead to an unstable environment, and ultimately, mass extinction.
Seattle, WA October 17. “BAHFest is a celebration of well-argued and thoroughly researched but completely incorrect scientific theory. Our brave speakers present their bad theories in front of a live audience and a panel of judges.” [$$]
The NSF Cyberlearning solicitation is out! Note that it has been substantially revised. For example, there is a new added focus for cyberlearning within the context of work at the human-technology frontier, and the Exploratory (EXP) category is no longer relevant as all proposals should be exploratory in nature. Proposal deadline is January 8, 2018.
The team, Autonomous Ballistics, took a divergent approach to nab the million-dollar award, focusing not on a smart gun, per se, but a smart holster, one that allows only the legal owner to draw the weapon. Their innovative design incorporates three different methods to release the gun from the holster. The first uses a fingerprint sensor to match the user’s fingerprints with the owner. The second is an RFID keycard that is worn by the user. Finally, the voice of the legal owner can also be used to authorize and free the gun.
Led by Sy Cohen, an NYU Tandon class of 2017 alumnus with a degree in mechanical engineering, the Autonomous Ballistics team includes Ashwin Raj Kumar, a PhD candidate in the Department of Mechanical and Aerospace Engineering; Jonathan Ng, a class of 2016 alumnus with a degree in mechanical engineering, and Eddilene Paola Cordero Pardo, an officer in the Colombian Navy and a Master’s degree student majoring in technology management. The team is mentored by NYU Tandon Adjunct Professor Anthony Clarke.
“Docker is a sophisticated software project built on the Go programming language for creating distributed and networked web applications, but it’s also wonderful for just accessing an Ubuntu bash terminal. To get started with Docker you should first install it (you want the Community Edition), then pull up your command line.”
Our two previous blog entries implied that there is a role games can play in driving the development of Reinforcement Learning algorithms. As the world’s most popular creation engine, Unity is at the crossroads between machine learning and gaming. It is critical to our mission to enable machine learning researchers with the most powerful training scenarios, and for us to give back to the gaming community by enabling them to utilize the latest machine learning technologies. As the first step in this endeavor, we are excited to introduce Unity Machine Learning Agents.
Maps are useful for understanding (very literally) the lay of the land, and guide our direct interactions with that land. But computers have also gotten really good at these tasks: GPS pinpoints our current location, geocoders can look up the street address, and navigation apps plan cross-country trips in a split second. Together, they assure us that we’ll never really be completely lost, and every destination is somehow reachable.
At the same time, our decision-making around places — where to go for lunch, if a certain day trip is feasible, and yes, whether a commute is practical — have shifted toward how long it takes to get there. But we still rely on traditional maps to do that: eyeballing straight-line distances, and running that through some alchemy of guesswork and firsthand experience to how long that trip would take.
What would a more directly useful visualization — indeed, a map of time — look like?