Posts Tagged ‘algorithms’
Supercomputer Reads the News to Successfully Forecast World Events
Kalev Leetaru of the University of Illinois determined that using the Nautilus SGI supercomputer to analyze news stories can help predict major world events. The analysis he used for the experiment was retrospective, feeding the computer millions of articles from which it was able to determine a deteriorating national sentiment towards Libya and Egypt before the revolutions in those countries. The system was also able to narrow down Osama Bin Laden's location to within 125 miles before he was found and killed last May.
More than 100 million articles were gathered for this study, from various sources including the New York Times archive, Open Source Center and BBC Monitoring (two organizations that monitor local media output worldwide). The system searched for two primary things in the articles: mood and location. Words such as “nice” or “horrible” were used to measure mood, and geocoding converted mentions of places such as “Cairo” or “Pakistan” to plottable coordinates.
For countries that experienced the “Arab Spring,” the supercomputer produced graphs that showed a noticeable decline in media sentiment both within each country and without. Before President Mubarak's resignation, the tone of media coverage of Egypt fell to one of its lowest points in 30 years, predicting something that U.S. government could not. As Leetaru told BBC news, the president's continued support of Mubarak showed that high-level analysis suggested Mubarak wasn't going anywhere. The graph, however, suggests otherwise.
Leetaru's next step is developing technology to allow this system to forecast major world events, rather than just analyzing them after the fact. He compares it to economic forecasting algorithms, as well as meteorology, in that none of those systems (including his) are perfect, but using them is far better than just guessing.
[BBC]
Supercomputer Reads the News to Successfully Forecast World Events
Kalev Leetaru of the University of Illinois determined that using the Nautilus SGI supercomputer to analyze news stories can help predict major world events. The analysis he used for the experiment was retrospective, feeding the computer millions of articles from which it was able to determine a deteriorating national sentiment towards Libya and Egypt before the revolutions in those countries. The system was also able to narrow down Osama Bin Laden's location to within 125 miles before he was found and killed last May.
More than 100 million articles were gathered for this study, from various sources including the New York Times archive, Open Source Center and BBC Monitoring (two organizations that monitor local media output worldwide). The system searched for two primary things in the articles: mood and location. Words such as “nice” or “horrible” were used to measure mood, and geocoding converted mentions of places such as “Cairo” or “Pakistan” to plottable coordinates.
For countries that experienced the “Arab Spring,” the supercomputer produced graphs that showed a noticeable decline in media sentiment both within each country and without. Before President Mubarak's resignation, the tone of media coverage of Egypt fell to one of its lowest points in 30 years, predicting something that U.S. government could not. As Leetaru told BBC news, the president's continued support of Mubarak showed that high-level analysis suggested Mubarak wasn't going anywhere. The graph, however, suggests otherwise.
Leetaru's next step is developing technology to allow this system to forecast major world events, rather than just analyzing them after the fact. He compares it to economic forecasting algorithms, as well as meteorology, in that none of those systems (including his) are perfect, but using them is far better than just guessing.
[BBC]
Video: Mabel the Robot Sets Speed Record For Bipedal Running
Few robots can run period, but only Mabel can run with such human-like qualities, according to Michigan researchers. Its weight is distributed like a human’s, with a heavy torso and light, flexible legs. Springs in the legs serve as tendons, allowing Mabel to bound like a real runner — it spends 40 percent of each stride in the air, while other running robots are more like speedwalkers, lifting off the ground for only 10 percent of each step.
The robot started off walking quickly over flat surfaces, and its programmers started improving the feedback algorithms that help it maintain its posture, according to a Michigan . Mabel does not quite run free, only moving attached to a metal bar like a horse at longeing.
Mabel’s programmers believe the robot’s realistic gait could be helpful for several applications, from powered prosthetic limbs to robotic exoskeletons. Or imagine a legion of robot runners that a human could ride, ostrich-style.
Home robots and rescue robots with a human stride could also be more effective than the cautious, two-step gait of other humanoids, according to Jessy W. Grizzle, who leads the lab where Mabel was built.
“If you would like to send in robots to search for people when a house is on fire, it probably needs to be able to go up and down stairs, step over the baby's toys on the floor, and maneuver in an environment where wheels and tracks may not be appropriate,” he said.
Mabel was built back in 2008 and researchers have been tweaking its design and programming. In the most recent tests, Mabel reached a top speed of 6.8 miles per hour, a pretty good clip. Watch it run below.
[ via ]
Algorithm Smooths 8-Bit Pixel Art Into Cute, Bubbly Vector Drawings
Johannes Kopf of Microsoft and Dani Lischinski of the Hebrew University built the algorithm and used it to modify the Super Mario World dolphin seen below, among other characters. Yoshi’s rounded eyes make him look drowsy, and the curvy Space Invaders somehow appear more angry.
The algorithm uses a combination of pixel analysis and spline curves to smooth out the drawings, the researchers explain. Other software programs can do this — Adobe Illustrator has its own vectorization process, as the blog Extreme Tech points out, and previous researchers have tried to smooth out pixel art.
But this algorithm works with 8-bit art, so it can be more specific, and the result is a more accurate portrayal. It figures out where the important connections lie between pixels, and then reshapes the pixel cells themselves so that they are connected to their neighbors via the cardinal (up and down, side to side) and diagonal edges. In the diagram below, you can see how the pixel cells around the ghost are reshaped from squares into asymmetric shapes.
The algorithm also minimizes spline curves as they pass through several points — rounding out the edges where the squares meet. As a result, staircasing artifacts that show up as blobs or weird halos in Adobe Trace are eliminated, and you get a shape that remains true to the artists’ original form, but without the characteristic blocky pixels.
“The best pixel art from the golden age of video games are masterpieces, many of which have become cultural icons that are instantly recognized by a whole generation, e.g. ‘Space Invaders’ or the 3-color Super Mario Bros. sprite,” Kopf and Lischinski write.
This algorithm casts them in a new light.
Practically speaking, this technology could improve inputs for modern display screens, which have a much higher refresh rate than older hardware for which these games were developed. Then browser-based video games would be even more fun.
[]
Financial Trading Algorithms Aren’t Just Making Deals, They’re Making War
It's an algorithm-on-algorithm free-for-all out there
Most trading algorithms execute simple tasks. Say a large institution wants to purchase a large chunk of stock in a company: the program will seek out shares and buy them in many, many small quantities so as not to send the price soaring with a massive order--and to stop other traders from seeing what they’re up to and getting in on the deal. Called VWAPs, they’re fairly benign.
But Donald MacKenzie, a professor of sociology at Edinburgh University, tells us that algorithms are now stalking the VWAPs, trying to surreptitiously figure out their intentions so they can get out in front of big trades, buying shares ahead of VWAPs and then selling them back to it at a gain.
But there are craftier and decidedly less ethical programs out there seeking to manipulate the market outright. Called “spoofers,” they might buy a big chunk of shares of a certain ticker. Then the program issues a bunch of buy orders that are fractionally below that price--an indicator to anyone watching the buy order volume that the stock is in demand. This would spur other algorithms (or human traders) to purchase the stock on the prospect that demand will push up the price. The spoofer then dumps its shares and quickly cancels its buy orders when the price rises on the buying pressure it created. Sneaky.
Basically, the algorithms are now trying to outfox other algorithms. That’s particularly troublesome given that computer programs are basically running financial markets these days. We all saw what happens when the algorithms get irrational during last year’s flash crash, when the entire U.S. market shed 6 percent of its value in five minutes due to algorithmic share dumping. While these programs are inherently stabilizing, it’s a bit frightening to think of what might happen if some super-algorithm really got the better of his computational counterparts.
[ via ]
By Teaching Computers ‘Regret,’ Engineers Hope to Teach Them to See the Future
While software may never know what it’s like to roll out of bed with splitting headache and dress quietly in the dark, it can certainly measure the distance between a desired outcome and the actual outcome achieved. And by doing so computers could learn to minimize “regret,” which in this case is measured by that distance.
TAU computer scientists working on learning theory and other thorny computer intelligence issues think that by teaching computers to reduce regret, they would essentially be teaching them to evaluate all the relevant variables surrounding an outcome in advance. This would allow them to do things like more efficiently route Internet traffic, prioritize server resource requests, or predict when traffic to a site might spike and provide the necessary capacity beforehand. And they could do it all based on data coming to them in real-time.
It could also do wonders for Google’s AdWords and Adsense businesses. Algorithms that can learn in real time and produce results with the least “regret” could sharpen ad targeting tools in a big way, turning Google’s desired outcome of higher ad revenues and the further trashing of their competitors into a likely actual outcome.
[]
New Scanner Tracks Zebras’ Built-In Bar Codes
A new algorithm can like they were bar codes, helping researchers track individual animals more easily.
“StripeSpotter,” designed at the University of Illinois-Chicago and Princeton University, will help researchers build biometric databases based on field photographs. The programmers are building a zebra-print database for Plains and Grevys zebras in Kenya.
users would only need a digital camera and a laptop capable of running the simple program. Take a picture of a zebra, and the StripeCode algorithm extracts certain image features, using a dynamic programming algorithm to compare them and search for a match. This way, ecologists can determine whether an animal has been observed before, and then take field notes, GPS coordinates and other information. If there is no match, the assumption is that the zebra has never been spotted before.
It could also be used for other striped animals like tigers and giraffes, the researchers say.
The system is described in a to be presented this month at the International Conference on Multimedia Retrieval.
[]