New Image-recognition Software Could Let Computers 'See' Like Humans Do

Source:

http://www.sciencedaily.com/releases/2008/05/080526000936.htm

ScienceDaily (May 26, 2008) — It takes surprisingly few pixels of information to be able to identify the subject of an image, a team led by an MIT researcher has found. The discovery could lead to great advances in the automated identification of online images and, ultimately, provide a basis for computers to see like humans do.

Antonio Torralba, assistant professor in MIT's Computer Science and Artificial Intelligence Laboratory, and colleagues have been trying to find out what is the smallest amount of information--that is, the shortest numerical representation--that can be derived from an image that will provide a useful indication of its content.
Deriving such a short representation would be an important step toward making it possible to catalog the billions of images on the Internet automatically. At present, the only ways to search for images are based on text captions that people have entered by hand for each picture, and many images lack such information. Automatic identification would also provide a way to index pictures people download from digital cameras onto their computers, without having to go through and caption each one by hand. And ultimately it could lead to true machine vision, which could someday allow robots to make sense of the data coming from their cameras and figure out where they are.
"We're trying to find very short codes for images," says Torralba, "so that if two images have a similar sequence [of numbers], they are probably similar--composed of roughly the same object, in roughly the same configuration." If one image has been identified with a caption or title, then other images that match its numerical code would likely show the same object (such as a car, tree, or person) and so the name associated with one picture can be transferred to the others.
"With very large amounts of images, even relatively simple algorithms are able to perform fairly well" in identifying images this way, says Torralba. He will be presenting his latest findings this June in Alaska at a conference on Computer Vision and Pattern Recognition. The work was done in collaboration with Rob Fergus at the Courant Institute in New York University and Yair Weiss of Hebrew University in Jerusalem.
To find out how little image information is needed for people to recognize the subject of a picture, Torralba and his co-authors tried reducing images to lower and lower resolution, and seeing how many images at each level people could identify.
"We are able to recognize what is in images, even if the resolution is very low, because we know so much about images," he says. "The amount of information you need to identify most images is about 32 by 32." By contrast, even the small "thumbnail" images shown in a Google search are typically 100 by 100.
Even an inexpensive current digital camera produces images consisting of several megapixels of data--and each pixel typically consists of 24 bits (zero or one) of data. But Torralba and his collaborators devised a mathematical system that can reduce the data from each picture even further, and it turns out that many images are recognizable even when coded into a numerical representation containing as little as 256 to 1024 bits of data.
Using such small amounts of data per image makes it possible to search for similar pictures through millions of images in a database, using an ordinary PC, in less than a second, Torralba says. And unlike other methods that require first breaking down an image into sections containing different objects, this method uses the entire image, making it simple to apply to large datasets without human intervention.
For example, using the coding system they developed, Torralba and his colleagues were able to represent a set of 12.9 million images from the Internet with just 600 megabytes of data--small enough to fit in the RAM memory of most current PCs, and to be stored on a memory stick. The image database and software to enable searches of the database, are being made publicly available on the web.
Of course, a system using drastically reduced amounts of information can't come close to perfect identification. At present, the matching works for the most common kinds of images. "Not all images are created equal," he says. The more complex or unusual an image is, the less likely it is to be correctly matched. But for the most common objects in pictures--people, cars, flowers, buildings--the results are quite impressive.
The work is part of research being carried out by hundreds of teams around the world, aimed at analyzing the content of visual information. Torralba has also collaborated on related work with other MIT researchers including William Freeman, a professor in the Department of Electrical Engineering and Computer Science; Aude Oliva, professor in the Department of Brain and Cognitive Sciences; and graduate students Bryan Russell and Ce Liu, in CSAIL. Torralba's work is supported in part by a grant from the National Science Foundation.
Torralba stresses that the research is still preliminary and that there will always be problems with identifying the more-unusual subjects. It's similar to the way we recognize language, Torralba says. "There are many words you hear very often, but no matter how long you have been living, there will always be one that you haven't heard before. You always need to be able to understand [something new] from one example."

Fausto Intilla - www.oloscience.com

Diamond-Like Crystals Discovered In Brazilian Beetle Solve Issue For Future Optical Computers

Source:

http://www.sciencedaily.com/releases/2008/05/080520090534.htm

ScienceDaily (May 21, 2008) — Researchers have been unable to build an ideal "photonic crystal" to manipulate visible light, impeding the dream of ultrafast optical computers. But now, University of Utah chemists have discovered that nature already has designed photonic crystals with the ideal, diamond-like structure: They are found in the shimmering, iridescent green scales of a beetle from Brazil.

"It appears that a simple creature like a beetle provides us with one of the technologically most sought-after structures for the next generation of computing," says study leader Michael Bartl, an assistant professor of chemistry and adjunct assistant professor of physics at the University of Utah. "Nature has simple ways of making structures and materials that are still unobtainable with our million-dollar instruments and engineering strategies."
The study by Bartl, University of Utah chemistry doctoral student Jeremy Galusha and colleagues is set to be published in a forthcoming edition of the journal Physical Review E.
The beetle is an inch-long weevil named Lamprocyphus augustus. The discovery of its scales' crystal structure represents the first time scientists have been able to work with a material with the ideal or "champion" architecture for a photonic crystal.
"Nature uses very simple strategies to design structures to manipulate light -- structures that are beyond the reach of our current abilities," Galusha says.
Bartl and Galusha now are trying to design a synthetic version of the beetle's photonic crystals, using scale material as a mold to make the crystals from a transparent semiconductor.
The scales can't be used in technological devices because they are made of fingernail-like chitin, which is not stable enough for long-term use, is not semiconducting and doesn't bend light adequately.
The University of Utah chemists conducted the study with coauthors Lauren Richey, a former Springville High School student now attending Brigham Young University; BYU biology Professor John Gardner; and Jennifer Cha, of IBM's Almaden Research Center in San Jose, Calif.
Quest for the Ideal or 'Champion' Photonic Crystal
Researchers are seeking photonic crystals as they aim to develop optical computers that run on light (photons) instead of electricity (electrons). Right now, light in near-infrared and visible wavelengths can carry data and communications through fiberoptic cables, but the data must be converted from light back to electricity before being processed in a computer.
The goal -- still years away -- is an ultrahigh-speed computer with optical integrated circuits or chips that run on light instead of electricity.
"You would be able to solve certain problems that we are not able to solve now," Bartl says. "For certain problems, an optical computer could do in seconds what regular computers need years for."
Researchers also are seeking ideal photonic crystals to amplify light and thus make solar cells more efficient, to capture light that would catalyze chemical reactions, and to generate tiny laser beams that would serve as light sources on optical chips.
"Photonic crystals are a new type of optical materials that manipulate light in non-classic ways," Bartl says. Some colors of light can pass through a photonic crystal at various speeds, while other wavelengths are reflected as the crystal acts like a mirror.
Bartl says there are many proposals for how light could be manipulated and controlled in new ways by photonic crystals, "however we still lack the proper materials that would allow us to create ideal photonic crystals to manipulate visible light. A material like this doesn't exist artificially or synthetically."
The ideal photonic crystal -- dubbed the "champion" crystal -- was described by scientists elsewhere in 1990. They showed that the optimal photonic crystal -- one that could manipulate light most efficiently -- would have the same crystal structure as the lattice of carbon atoms in diamond. Diamonds cannot be used as photonic crystals because their atoms are packed too tightly together to manipulate visible light.
When made from an appropriate material, a diamond-like structure would create a large "photonic bandgap," meaning the crystalline structure prevents the propagation of light of a certain range of wavelengths. Materials with such bandgaps are necessary if researchers are to engineer optical circuits that can manipulate visible light.
On the Path of the Beetle: From BYU to Belgium and Brazil
The new study has its roots in Richey's science fair project on iridescence in biology when she was a student at Utah's Springville High School. Gardner's group at BYU was helping her at the same time Galusha was using an electron microscope there and learned of Richey's project.
Richey wanted to examine an iridescent beetle, but lacked a complete specimen. So the researchers ordered Brazil's Lamprocyphus augustus from a Belgian insect dealer.
The beetle's shiny, sparkling green color is produced by the crystal structure of its scales, not by any pigment, Bartl says. The scales are made of chitin, which forms the external skeleton, or exoskeleton, of most insects and is similar to fingernail material. The scales are affixed to the beetle's exoskeleton. Each measures 200 microns (millionths of a meter) long by 100 microns wide. A human hair is about 100 microns thick.
Green light -- which has a wavelength of about 500 to 550 nanometers, or billionths of a meter -- cannot penetrate the scales' crystal structure, which acts like mirrors to reflect the green light, making the beetle appear iridescent green.
Bartl says the beetle was interesting because it was iridescent regardless of the angle from which it was viewed -- unlike most iridescent objects -- and because a preliminary electron microscope examination showed its scales did not have the structure typical of artificial photonic crystals.
"The color and structure looked interesting," Bartl says. "The question was: What was the exact three-dimensional structure that produces these unique optical properties?"
The Utah team's study is the first to show that "just as atoms are arranged in diamond crystals, so is the chitin structure of beetle scales," he says.
Galusha determined the 3-D structure of the scales using a scanning electron microscope. He cut a cross section of a scale, and then took an electron microscope image of it. Then he used a focused ion beam -- sort of a tiny sandblaster that shoots a beam of gallium ions -- to shave off the exposed end of the scale, and then took another image, doing so repeatedly until he had images of 150 cross-sections from the same scale.
Then the researchers "stacked" the images together in a computer, and determined the crystal structure of the scale material: a diamond-like or "champion" architecture, but with building blocks of chitin and air instead of the carbon atoms in diamond.
Next, Galusha and Bartl used optical studies and theory to predict optical properties of the scales' structure. The prediction matched reality: green iridescence.
Many iridescent objects appear that way only when viewed at certain angles, but the beetle remains iridescent from any angle. Bartl says the way the beetle does that is an "ingenious engineering strategy" that approximates a technology for controlling the propagation of visible light.
A single beetle scale is not a continuous crystal, but includes some 200 pieces of chitin, each with the diamond-based crystal structure but each oriented a different direction. So each piece reflects a slightly different wavelength or shade of green.
"Each piece is too small to be seen individually by your eye, so what you see is a composite effect," with the beetle appearing green from any angle, Bartl explains.
Scientists don't know how the beetle uses its color, but "because it is an unnatural green, it's likely not for camouflage," Bartl says. "It could be to attract mates."
The study was funded by the National Science Foundation, American Chemical Society, the University of Utah and Brigham Young University.

Fausto Intilla - www.oloscience.com

Researchers teach 'Second Life' avatar to think

Source:

http://www.lse.co.uk

TROY, N.Y. (AP) - Edd Hifeng barely merits a second glance in 'Second Life.' A steel-gray robot with lanky limbs and linebacker shoulders, he looks like a typical avatar in the popular virtual world.But Edd is different.His actions are animated not by a person at a keyboard but by a computer.
Edd is a creation of artificial intelligence, or AI, by researchers at Rensselaer Polytechnic Institute, who endowed him with a limited ability to converse and reason. It turns out 'Second Life' is more than a place where pixelated avatars chat, interact and fly about. It's also a frontier in AI research because it's a controllable environment where testing intelligent creations is easier.'It's a very inexpensive way to test out our technologies right now,' said Selmer Bringsjord, director of the Rensselaer Artificial Intelligence and Reasoning Laboratory.Bringsjord sees Edd as a forerunner to more sophisticated creations that could interact with people inside three-dimensional projections of settings like subway stops or city streets. He said the holographic illusions could be used to train emergency workers or solve mysteries.But first, a virtual reality check.Edd is not running rampant through the cyber streets of 'Second Life.' He goes only where Bringsjord and his graduate students place him for tests. He can answer questions like 'Where are you from?' but understands only English that has previously been translated into mathematical logic.'Second Life' is attractive to researchers in part because virtual reality is less messy than plain-old reality. Researchers don't have to worry about wind, rain or coffee spills.And virtual worlds can push along AI research without forcing scientists to solve the most difficult problems -- like, say, creating a virtual human -- right away, said Michael Mateas, a computer science professor at the University of California at Santa Cruz.Researching in virtual realities has become increasingly popular the past couple years, said Mateas, leader of the school's Expressive Intelligence Studio for AI and gaming.'It's a fantastic sweet spot -- not too simple, not too complicated, high cultural value,' he said.Bringsjord is careful to point out that the computations for Edd's mental feats have been done on workstations and are not sapping 'Second Life' servers. The calculations will soon be performed on a supercomputer at Rensselaer with support from research co-sponsor IBM Corp.Operators of 'Second Life' don't seem concerned about synthetic agents lurking in their world. John Lester, Boston operations manager for Linden Lab, said the San Francisco-based company sees
a 'fascinating' opportunity for AI to evolve.
'I think the real future for this is when people take these AI-controlled avatars and let them free in 'Second Life,'' Lester said, ' ... let them randomly walk the grid.'That is years off by most experts' estimations. Edd's most sophisticated cognitive feat so far -- played out in 'Second Life' and posted on the Web -- involves him witnessing a gun being switched from one briefcase to another. Edd was able to infer that another 'Second Life' character who left the room during the switch would incorrectly think the gun was still in the first suitcase.This ability to make inferences about the thoughts of others is significant for an AI agent, though it puts Edd on par with a 4-year-old -- and the calculus required 'under the hood' to achieve this feat is mind-numbingly complex.A computer program smart enough to fool someone into thinking they're interacting with another person -- the traditional Holy Grail for AI researchers -- has been elusive. One huge problem is getting computers to understand concepts imparted in language, said Jeremy Bailenson, director of the Virtual Human Interaction Lab at Stanford University.AI agents do best in tightly controlled environments: Think of automated phone programs that recognize your responses when you say 'operator' or 'repair.'Bringsjord sees 'Second Life' as a way station. He eventually wants to create other environments where more sophisticated creations could display courage or deceive people, which would be the first step in developing technology to detect deception.The avatars could be projected at RPI's $145 million Experimental Media and Performing Arts Center, opening in October, which will include spaces for holographic projections. Officials call them 'holodecks' in homage to the virtual reality room on the 'Star Trek' television series.That sort of visual fidelity is many years down the line, just like complex AI. John Kolb, RPI's chief information officer, said the best three-dimensional effects still require viewers to wear special light-polarizing glasses.'If you want to do texture mapping on a wall for instance, that's easy. We can do that today,' Kolb said. 'If you want to start to build cognitive abilities into avatars, well, that's going to take a bit more work.'

Fausto Intilla - www.oloscience.com

Braille Converter Bridges The Information Gap

Source:

http://www.sciencedaily.com/releases/2008/05/080508174310.htm

ScienceDaily (May 12, 2008) — A free, e-mail-based service that translates text into Braille and audio recordings is helping to bridge the information gap for blind and visually impaired people, giving them quick and easy access to books, news articles and web pages.
Developed by European researchers, the RoboBraille service offers a unique solution to the problem of converting text into Braille and audio without the need for users to operate complicated software.
“We started working in this field 20 years ago, developing software to translate text into Braille, but we discovered that users found the programs difficult to use – we therefore searched for a simpler solution,” explains project coordinator Lars Ballieu Christensen, who also works for Synscenter Refsnaes, a Danish centre for visually impaired children.
The result of the EU-funded project was RoboBraille, a service that requires no more skill with a computer than the ability to send an e-mail.
Users simply attach a text they want to translate in one of several recognised formats, from plain text and Word documents to HTML and XML. They then e-mail the text to the service’s server. Software agents then automatically begin the process of translating the text into Braille or converting it into an audio recording through a text-to-speech engine.
“The type of output and the language depends on the e-mail address the user sends the text to,” Christensen says. “A document sent to .org would be converted into spoken British English while a text sent to .org would be translated from Portuguese into six-dot Braille.”
The user then receives the translation back by e-mail, which can be read on a Braille printer or on a tactile display, a device connected to the computer with a series of pins that are raised or lowered to represent Braille characters.
RoboBraille can currently translate text written in English, Danish, Italian, Greek and Portuguese into Braille and speech. The service can also handle text-to-speech conversions in French and Lithuanian.
Christensen notes that the RoboBraille partners are constantly working on adding new languages to the service and plan to start providing Braille and audio translations for Russian, Spanish, German and Arabic. They are also working on making the service compatible with PDF documents and text scanned from images.
Up to 14,000 translations a day
At present, the service translates an average of 500 documents a day, although it could handle as many as 14,000. RoboBraille can return a simple text in Braille in under a minute while taking as long as 10 hours to provide an audio recording of a book.
As of January, the RoboBraille system had carried out 250,000 translations since it first went online.
The team have won widespread recognition for their work, receiving the 2007 Social Contribution Award from the British Computer Society in December while in April they were awarded the 2008 award for technological innovation from Milan-based Well-Tech.
“We initially started offering the service only in Denmark but to make it viable commercially we needed to broaden our horizons. Hence the eTen project which allowed us to involve other organisations across Europe in developing and expanding the service, not only geographically but also in terms of users,” Christensen says.
In addition to the blind and visually impaired, the service can also help dyslexics, people with reading difficulties and the illiterate. The project partners plan to continue to offer the service for free to such users and other individuals, while in parallel developing commercial services for companies and public institutions.
“Pharmaceutical companies in Europe will soon be required to ensure all medicine packaging is labelled in Braille and we are currently working with three big firms to provide that service,” Christensen explains. “Banks and insurance companies are also interested in using it to provide statements in Braille as too is the Danish tax office. In Italy there is interest in using it in the tourism sector.”
The RoboBraille team, which recently received an €1.1 million grant over four years from the Danish government, expect the service to be profitable within four or five years.
And although they are not actively seeking investors, they are interested in partnerships with organisations interested in collaborating on specific social projects.
RoboBraille was funded under the EU's eTEN programme for market validation and implementation.

Fausto Intilla - www.oloscience.com

Quantum Cryptography Cracked?

Source:

http://www.spectrum.ieee.org/apr08/6203

29 April 2008—Quantum cryptography, touted by scientists as the ultimate unbreakable code, may turn out to be susceptible to eavesdropping after all when implemented practically, according to a Swedish duo.
“Quantum codes are supposed to guarantee 100 percent security,” says Jan-Ake Larsson, associate professor of mathematics at Linkoeping University, in Sweden. “If they don't live up to that promise, that's a problem.”
Larsson and his former graduate student Jorgen Cederlof, who now works for Google, say they have spotted a flaw in practical quantum codes. Their report on this flaw and a patch for the problem appear in the April issue of the IEEE Transactions on Information Theory.
The most secure codes currently in use rely on public-key cryptography, whose security stems from the fact that computers today cannot factor very large numbers within a useful time period. However, in theory, given sufficiently powerful computers, these codes can be cracked.
Quantum cryptography, in contrast, is supposed to be unbreakable, even in theory, because its security is based on a fundamental tenet of quantum mechanics. It turns out that the very act of measurement in quantum mechanics changes the nature of the quantum system being observed. Thus, if an eavesdropper listens in on a quantum message between two parties, he or she changes the message in a way that is detectable. Through a multistep process, quantum encryption systems—and there are at least three on the market now—use the security of quantum mechanics to generate cryptographic keys. These quantum keys are ciphers used to encode and decode messages.
The process of key generation, though based on quantum physics, also requires exchanging some information on a regular “classical” channel. Eavesdropping on the classical channel cannot be detected. One of the final steps in setting up a quantum key is to authenticate the communicating parties—determining that Bob is really talking to Alice, not some eavesdropper.
If there is no authentication, Alice and Bob will be open to a “man in the middle” attack, as it is termed by code breakers. The attack would work like this, Cederlof explains: “Now Eve comes along, buys a couple of [quantum encryption] devices identical to the ones Alice and Bob have, cuts the cables between Alice and Bob, and connects her devices at both ends. Now Alice will think she is talking to Bob, but in reality she is talking to Eve. Eve just acts as Bob would have, and after a while Alice and Eve have created a shared secret key. The same thing happens between Eve and Bob. When Alice tries to send an encrypted message to Bob, she will encrypt it with a key known only to Eve (but which Alice thinks only Bob knows). Eve intercepts the message, decrypts it, reads it, encrypts it with the key she shares with Bob, and sends it to Bob. Alice and Bob never suspect anything.”
The way around this is to communicate classically and make sure Alice is really talking to Bob. But that is exactly where the vulnerability lies.
“To our surprise, the authentication was not secure,” says Larsson. He and Cederlof say that it is difficult to eavesdrop, but the possibility does exist. In their paper they suggest a patch. “The modification we propose is basically an extra exchange of a small amount of random bits on the classical channel,” says Larsson.
According to Tassos Nakassis a computer scientist at the National Institute of Standards and Technology (NIST), in Gaithersburg, Md., the error may have originated because quantum cryptography is an emerging interdisciplinary field that combines advanced quantum physics with traditional code making. Authentication and its weaknesses may have gotten lost in the conversation between quantum physicists and classical cryptographers.
The Swedes went looking in just the right place for a vulnerability, according to Bruce Schneier, an expert in cryptography and chief technology officer at BT Counterpane, in Santa Clara, Calif. “Authentication has always been a problem with quantum crypto,” he says.
Audrius Berzanskis, chief operating officer at the quantum cryptography systems firm MagiQ Technologies, in New York City, claims his firm's systems are immune to this kind of attack, because they are overly conservative with respect to how they treat errors in the quantum channel—whether or not the errors are caused by an eavesdropper. This conservatism comes at the cost of the rate at which quantum keys are generated. And Berzanskis adds that Larsson and Cederlof's patch might allow the key rate to increase. Experts from outside quantum cryptography companies agree that the vulnerability is real, but most think it would be impractical to exploit.
“This is an interesting issue and worthy of the awareness of the community,” says physicist Joshua Bienfang, who works on quantum cryptography at NIST. But he notes that Larsson and Cederlof correctly emphasize that the attack relies on Eve capitalizing on opportunities that occur with very low probability. In their worst-case scenario, with a computationally omnipotent Eve, they estimate it would take something on the order of nine months to break the system. And he says that the patch offered should “firmly shut the door on this type of attack.”
Norbert Lutkenhaus, a physicist at the Institute for Quantum Computing, in Canada, summed it up. “Practically, I don't think it is a threat of any kind,” he says. “But it is good to know about the vulnerability.”

Fausto Intilla - www.oloscience.com