Web Concepts

Posts Tagged ‘computer science’

Dassault Systemes, the french software developer, has built a computer model of Mougin’s idea. And after 15 engineers ran the problem through their models, they found that the idea is more or less perfectly feasible. Towing an iceberg from somewhere around Newfoundland to the northwest coast of Africa would only take around five months and could still retain more than 60 percent of the iceberg’s mass. The downside: it would cost about $10 million.

The simulation accounted for the costs associated with fitting a huge insulating sleeve around a seven-ton iceberg, towing it across the Atlantic via tugboat and kite-sail (at a speed of about one knot), and then distributing the water inland from the coast. A real-world attempt would likely call for a much larger iceberg--a 30-ton iceberg could keep half a million people in drinking water for a year--but the fundamentals, the model says, are sound.

Since some 40,000 icebergs break away from the polar ice caps each year anyhow, it would make productive use of the fact that we’re all slowly drowning (though few of those ‘bergs are big enough to be worth towing). And it could help deal with serious drought-related human disasters like the one currently unfolding in Somalia.

For his part Mougin is reportedly re-energized by models. At 86, he’s raising funds for a real-world attempt at making his idea a reality.

[PhysOrg]

In all seriousness, the problem of “infinite loops” is beyond annoying. It saps productivity from software (and those using it). Infinite loops occur when a program gets stuck executing a single block of code over and over again (you probably know this as “freezing” or “f*&k!”). They often occur during functions where a program is trying to perform a task on many pieces of data in sequence, like when it searches for a word in a document for instance.

The problem occurs when the program, for whatever reason, doesn’t know when to stop repeating that operation, or executing that same segment of code repeatedly. Hence the term loop. And hence your frustration, as now your program won’t let you do anything else, including save your progress. So MIT researchers built a sort of crash cart for frozen programs that can shake them from an infinite loop, moving them along to the next logical block of code.

The program, appropriately named Jolt, recognizes infinite loops by examining the program’s use of memory. Say your program appears to be stalled. When you run Jolt, it takes a look at the program’s memory after each repetition of that loop. If there’s a change after each execution, your program is probably doing something useful. If not, it’s simply hung up in an infinite loop. Jolt then looks for the first instruction that follows the code the program is stuck on and forces the program to move ahead (for you programming types who are interested, there’s a much more thorough nuts-and-bolts description over at MIT News).

That forced procedure may not restore the program to full functionality--for instance, Jolt (and its binary cousin Bolt) may not push the program to the correct next instruction--but ideally it will at least put the program in a state where you can save, quit, and relaunch. That beats retyping your term paper. After all, you stayed up all night just to get it finished.

[MIT News]

The material is GST--so named for the materials it contains (germanium, antimony, and tellurium--and it possesses just the kind of phase-changing properties that researchers are looking for. Phase-change alloys like GST can exist in various phases, ranging from crystalline, ordered structures to chaotic, amorphous structures. Their states depends on various external factors, like heat or charge.

In DVDs, GST allows discs to be embedded with binary ones and zeros that can later be read by a laser. That’s fine for film screening purposes, but GST can actually exist in various degrees of phase change between completely crystalline and completely amorphous, and thus it can store information across a wider range of values--just like a neuron, which fires only when a build-up of incoming signals reaches a certain threshold.

The GST neurons/synapses developed by the University of Exeter and Stanford team also can adjust the strength of the synapses between them--a key characteristic of inter-neuron communication--because of its inherent ability to modify its electrical resistance. This allows the GST neurons to adjust the strength of the connections between them to signify the importance of incoming signals and prioritize signals flowing through a neural network.

All said, those two qualities--the ability to store information across a range of phase states, and a low-power, adjustable-strength synapse--make for a pretty nice electronic analog for a working brain. But the work is very preliminary, and it’s one thing to have a few working synthetic neurons in the lab, and quite another to have a working network of thousands or millions (much less hundreds of millions).

In other words, it’s pretty amazing that DVD tech can do these things, but a proper brain-like computer is still many years and several breakthroughs distant.

[New Scientist]

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.

[American Friends of Tel Aviv University]

Memristors are the fourth fundamental piece of electronic circuitry after capacitors, inductors, and resistors. They are basically resistors that remember what state of resistance they were in the last time a current passed through them. They were proposed more than three decades ago but were only first created in recent years by HP. And they are expected to drive revolutionary advances in future electronics for several reasons, not least of which is their ability to behave more like neurons than conventional electronic circuits.

Which brings us back to mazes. A maze can have varying degrees of complexity. It might have several solutions. A conventional processor would solve a maze the same way a person might--start at the starting point and weave itself through, noting dead ends and retrying until it gets all the way through to the end. Depending on the complexity of the maze, this could take some time.

To demonstrate how memristors could do better, Yuriy Pershin at the University of South Carolina and Massimiliano Di Ventra at UCSD constructed a sort of universal maze out of a grid of memristors which could be adopted to reflect any maze by disabling certain connections between memristors through which electricity cannot flow. Using this memristor array, once any maze design is imposed on their processor it can be solved for by simply applying voltage to the maze entrance and a ground to the finish.

That doesn’t sound so groundbreaking until you think it through. The memristor grid, unlike a conventional computer program, actually works in parallel, with all of the memristors working on solving the maze simultaneously. If there are multiple solutions, those are solved for simultaneously as well. Beyond that, the memristors will “remember” the solution(s) in their states for recall or use later.

That’s not such a big deal if you’re just solving a maze, but if you’re applying this power to robotics, graph theory, network optimization, or a slew of other computing models, it has the power to work much faster and more efficiently than computing complex problems in series.

Perhsin and Di Ventra’s processor amounts to the first application of memristor networks to massively parallel computing. Considering it such computing more closely mirrors the way the human brain works--or the way a brain-like computer might work--solving a maze with a memristor is about thrilling solving a simple puzzle can get.

[Technology Review]

Yet another reason to save them from extinction

Bumblebees can solve the classic "traveling salesman" problem, which keeps supercomputers busy for days. They learn to fly the shortest possible route between flowers even if they find the flowers in a different order, according to a new British study.

The traveling salesman problem is an http://en.wikipedia.org/wiki/NP-hardNP-hard (read: very hard) problem in computer science; it involves finding the shortest possible route between cities, visiting each city only once. Bees are the first animals to figure this out, according to Queen Mary University of London researchers.

Bees need lots of energy to fly, so they seek the most efficient route among networks of hundreds of flowers. They navigate using angles of sunlight, which helps them find their way home, researchers say. To do this, their tiny brains must pack a powerful memory.(Old bees are more forgetful, according to a separate study that came out last week.)

To test bee problem-solving, researchers Lars Chittka and Mathieu Lihoreau tested bees’ response to computer-controlled artificial flowers. They wanted to see whether the bees would go after the flowers in the order in which they were discovered, or if they would figure out the shortest route among all the flowers even as new ones were added. The bees explored the locations of the flowers and quickly figured out the shortest paths among them, according to a Queen Mary news release.

This is no small feat, especially considering bee brains are about as big as a microdot. When it comes to intelligence, size apparently does not matter.

Earlier this year, researchers showed that bees recognize individual faces because they can make out the relative patterns that make up a face. The new research further suggests bees are highly sophisticated problem solvers, and that better understanding of their brains could improve our understanding of network problems like traffic flows, supply chains and epidemiology.

The research will be published this week in the journal The American Naturalist.

[Queen Mary University of London]

Researchers at a Rice University lab demonstrated last year that current could repeatedly break and reconnect tiny, 10-nanometer graphite strips to create reliable, very small memory bits. At the time they didn’t understand why the graphite did this so well; now, grad student Jun Yao has figured it out, and it has little to do with graphite.

Using silicon oxide, an insulator, as the meat in a tiny semiconductor sandwich, Yao showed that the electrodes will strip oxygen from the silicon oxide leaving behind a small chain of nano-sized silicon crystals. That crystal chain can then be connected or broken repeatedly by varying the electrical charge passed through it, creating a tiny switch that is always either on or off. And by tiny, we mean very tiny; Yao’s silicon oxide switch is just five nanometers (that’s five billionths of a meter) wide.

The graphite switches that seemed impressive last year were double that size, and conventional electronics can’t even come close to switches that small. Flash memory, in theory, will bottom out at 20 nanometers. Other conventional pathways might someday hit 10 nanometers, but it will be expensive to get there. Silicon oxide is already used in chip manufacturing and would be relatively easy to integrate into existing chip manufacturing tech.

Moreover, unlike flash memory silicon-oxide chips wouldn’t need to hold a charge and it’s perfectly suited to be arrayed in 3-D structures that can further help cram more switches onto a given switch, meaning chips get more memory for every nanometer of real estate. An Austin tech company is already testing a 1,000 memory element chip in collaboration with Yao and his colleagues at Rice. If the technology doesn’t hit any serious obstacles, single chips with memory comparable to today’s high-capacity disk drives could be a reality in just five years.