U.S. Army researchers, in cooperation with university scientists, have developed new techniques to train robots or computer programs to perform tasks under the guidance of a human instructor. They want to use deep learning, a class of machine learning algorithms that are loosely inspired by the brain, to train a robot to learn how to perform tasks by viewing video streams in a short amount of time with a human trainer. The researchers from the U.S. Army Research Laboratory and the University of Texas at Austin developed a new algorithm called Deep TAMER, an extension of Training an Agent Manually via Evaluative Reinforcement (TAMER). Their findings, which were released Friday, will be presented to an academic conference of the Association for the Advancement of Artificial Intelligence in New Orleans, Louisiana on Feb. 2-7. In their earlier work, the researchers taught a robot to behave in a situation similar to the way a dog was trained to do a trick.

The colored lines represent calculated particle tracks from particle collisions occurring within Brookhaven National Laboratory's STAR detector at the Relativistic Heavy Ion Collider, and an illustration of a digital brain. The yellow-red glow at center shows a hydrodynamic simulation of quark-gluon plasma created in particle collisions. Computers can beat chess champions, simulate star explosions, and forecast global climate. We are even teaching them to be infallible problem-solvers and fast learners. And now, physicists at the Department of Energy's Lawrence Berkeley National Laboratory (Berkeley Lab) and their collaborators have demonstrated that computers are ready to tackle the universe's greatest mysteries.

For more than 30 years, Geoffrey Hinton hovered at the edges of artificial intelligence research, an outsider clinging to a simple proposition: that computers could think like humans do--using intuition rather than rules. The idea had taken root in Hinton as a teenager when a friend described how a hologram works: innumerable beams of light bouncing off an object are recorded, and then those many representations are scattered over a huge database. Hinton, who comes from a somewhat eccentric, generations-deep family of overachieving scientists, immediately understood that the human brain worked like that, too--information in our brains is spread across a vast network of cells, linked by an endless map of neurons, firing and connecting and transmitting along a billion paths. He wondered: could a computer behave the same way? The answer, according to the academic mainstream, was a deafening no. Computers learned best by rules and logic, they said. And besides, Hinton's notion, called neural networks--which later became the groundwork for "deep learning" or "machine learning"--had already been disproven. In the late '50s, a Cornell scientist named Frank Rosenblatt had proposed the world's first neural network machine. It was called the Perceptron, and it had a simple objective--to recognize images. The goal was to show it a picture of an apple, and it would, at least in theory, spit out "apple." The Perceptron ran on an IBM mainframe, and it was ugly.

Microsoft plans to significantly expand its Montreal research lab and has hired a renowned artificial intelligence expert, Geoffrey Gordon, to be the lab's new research director. The company said Wednesday that it hopes to double the size of Microsoft Research Montreal within the next two years, to as many as 75 technical experts. The expansion comes as Montreal is becoming a worldwide hub for groundbreaking work in the fields of machine learning and deep learning, which are core to AI advances. "Montreal is really one of the most exciting places in AI right now," said Jennifer Chayes, a technical fellow and managing director of Microsoft Research New England, New York City and Montreal. In a meeting at the World Economic Forum in Davos, Canadian Prime Minister Justin Trudeau and Microsoft CEO Satya Nadella discussed Microsoft's ongoing investment in Canada and the expansion of the Montreal lab, including Gordon's hiring.

Cyber attacks have become more sophisticated, happen faster and cause more business disruption than ever before. Preventative tools like anti-virus and IDS, have not kept pace. This is a problem, because if we are not going to make any progress at all in defending against cyber attacks we will need both automation and artificial intelligence to help win the fight. For far too long, the key strategy in defending networks has been prevention, or not allowing bad things to happen in the first place. This strategy involves defending every single ingress and egress point and protecting against all threats to these points at all times.

This presentation was recorded at #H2OWorld 2017 in Mountain View, CA. He is working on large-scale and interpretable machine learning applied to natural language processing. Sameer was a Postdoctoral Research Associate at the University of Washington and received his PhD from the University of Massachusetts, Amherst, during which he also worked at Microsoft Research, Google Research, and Yahoo! He was awarded the Adobe Research Data Science Faculty Award, was selected as a DARPA Riser, won the grand prize in the Yelp dataset challenge, and received the Yahoo! Sameer has published extensively at top-tier machine learning and natural language processing conferences.

Among those presenting at last week's World Economic Forum in Davos, Switzerland, was Mary "Missy" Cummings, a Duke professor in the Department of Mechanical Engineering and Materials Science whose areas of expertise include artificial intelligence. Duke Today asked Cummings, who has presented before at the forum, about this year's event. Q: How, if at all, was the mood at Davos different this year than in previous years in which you have attended? CUMMINGS: Interestingly I did not see any real difference in the overall mood over last year, but (President) Trump's presence certainly created a buzz that was not as palpable as last year. Q: What were the 3 major points that you made during your presentation this year?

Earlier this week, the mayor of Louisville, Kentucky told reporters that he wants the city to field a fleet of drones that automatically survey areas after guns are fired. The city would detect firearm discharges using its existing ShotSpotter system, WDRB reported, and immediately send the UAVs to the scene, potentially before emergency responders are even called. But this isn't coming out of nowhere: Louisville could just be the first of over 300 cities that have applied to a federal program that provides funding for local governments that are trying to start their own drone programs. Cities had to apply for the FAA and DOT's US Unmanned Aerial System Integration Pilot Program by the end of last November, but of the hundreds of applicants, only five will be chosen. So far, only Louisville is proposing this particular use for a drone fleet, according to Gizmodo. But the city's mayor and civic innovation chief believe a host of UAVs buzzing in to photograph or video record a location and leaving thereafter would be less of a privacy violation than blanketing the city in security cameras -- and be cheaper, too.

UK-based legal AI company, Luminance, is to open an office in Singapore to meet increasing demand in the Asia-Pacific for doc review automation. The fast-growing legal AI venture said that this move follows winning several clients in Singapore and Australia. Luminance has also released Version 3.0 of its contract review technology and bagged UK law firm, Bird & Bird, as a new client. Bird & Bird would appear to be the second UK law firm to publicly announce it is using Luminance. The other is Slaughter and May, which owns a financial stake in the company. Although, it is understood several other UK firms are currently piloting the AI platform.

We present a new modeling technique for solving the problem of ecological inference, in which individual-level associations are inferred from labeled data available only at the aggregate level. We model aggregate count data as arising from the Poisson binomial, the distribution of the sum of independent but not identically distributed Bernoulli random variables. We relate individual-level probabilities to individual covariates using both a logistic regression and a neural network. A normal approximation is derived via the Lyapunov Central Limit Theorem, allowing us to efficiently fit these models on large datasets. We apply this technique to the problem of revealing voter preferences in the 2016 presidential election, fitting a model to a sample of over four million voters from the highly contested swing state of Pennsylvania. We validate the model at the precinct level via a holdout set, and at the individual level using weak labels, finding that the model is predictive and it learns intuitively reasonable associations.