algorithm
Latest
Facebook says it 'can do better' with its 'Year in Review' slideshows
Facebook inadvertently opened a lot of old wounds with its automated "Year in Review" slideshow feature. While it was meant to highlight people's favorite moments, it also reminded many of deaths, divorces and other tragedies that they tried to leave in the past. The company is clearly ready to own up to its mistake, though. Product manager Jonathan Gheller has apologized to Eric Meyer (whose story about his daughter's death drew attention to the problem) for the insensitivity and declared that the "Year in Review" team "can do better" in the future. While it's not clear what those changes will involve, The Guardian notes that Facebook has already changed the ending of the slideshow from "it's been a great year" to "see you next year" to avoid making presumptions.
Facebook's 'Year In Review' shows tragic side of software's shortcomings
Facebook's automated 'year in review' slideshows are meant to surface highlights from the year that was, but for some the virtual scrapbook simply brings back bad memories. In the case of web designer Eric Meyer, a photo of his recently deceased daughter appeared, surrounded by confetti, illustrations of party goers dancing and the exclamation "Eric, here's what your year looked like!" In response, Eric wrote a blog post about what he refers to as that app's "Inadvertent Algorithmic Cruelty," and pointing to the shortcomings of modern software design. While many have complained of the relentless onslaught of ads for automated journals like these, for people like Meyer, the persistence isn't just an annoyance.
Computers are ranking the world's important authors
Trying to rate the world's literary giants is tricky at best. Do you go by the number of books sold? The long-term cultural impact? If you're Dartmouth College researcher Allen Riddell, you make computers decide. As part of an effort to determine which books would be most valuable in the public domain, Riddell has developed an author ranking algorithm that determines the most important authors who died in a given year. The system ranks writers based on the age, length and popularity of their Wikipedia articles, along with the number of titles they have in the public domain. If an author gets a lot of attention but doesn't have many freely available works, that person climbs the charts and is more likely to have titles published on free literature sites like Project Gutenberg.
Computers are learning to size up neighborhoods using photos
Us humans are normally good at making quick judgments about neighborhoods. We can figure out whether we're safe, or if we're likely to find a certain store. Computers haven't had such an easy time of it, but that's changing now that MIT researchers have created a deep learning algorithm that sizes up neighborhoods roughly as well as humans. The code correlates what it sees in millions of Google Street View images with crime rates and points of interest; it can tell what a sketchy part of town looks like, or what you're likely to see near a McDonald's (taxis and police vans, apparently).
Twitter CFO hints at big changes to how your timeline works
Twitter has already started to look more like Facebook, and it might soon start acting more like it too. You see, company CFO Anthony Noto hinted yesterday that the reverse-chronological firehose of tweets that some users hold so dear may give way to a more curated collection of messages cast into the digital ether. To hear him tell it, the Twitter experience as we know it "isn't the most relevant" to the people who actually use the service (a notion that some people would definitely disagree with). That tidbit was lodged inside a broader conversation (which the Wall Street Journal captured) about improving Twitter's search functionality -- Noto pointed out the need for "an algorithm that delivers the depth and breadth of the content we have on a specific topic and then eventually as it relates to people." Those last few words seem crucial -- it sounds like he wants the Twitter experience to become one where content is tailored and presented differently depending on how relevant it is to the user. In the end, it might wind up getting Twitter a bunch of new users (which is exactly what all those antsy shareholders want to see), but would it really be worth alienating the service's hardcore fans?
A sheep dog's herding instinct may teach robots a lesson in crowd control
Here's a noodle-scratcher to occupy yourself with for a few moments: what makes a sheep dog so darned good at rounding up the woolly ruminants they're named after? A possible answer - according to The Telegraph, researchers at Swansea University believe those dogs are constantly searching for and minimizing the gaps between the sheep before it herds them all forward. What's the big deal? Well, those very same researchers think that behavior can be boiled down into an algorithm that could be used to (among other things) program robots to replace those savvy canines. Sure, some old-school shepherds may scoff, but using awkward-looking machines to round up livestock isn't exactly new territory. And if a robot can "understand" how to steer some relatively dumb animals around a field, it stands to reason that logic could be used to guide other organisms around... like humans trying to escape a burning building, for instance. No, really! Swansea University's Dr. Andrew King says there's a whole host of ways to adapt that animal knowledge into robotic know-how, like "crowd control, cleaning up the environment, herding of livestock, [and] keeping animals away from sensitive areas".
Cars may soon know when you're on the phone behind the wheel
Word to the wise, kids: do not muck around with your phone while driving. Some of you probably won't be able to help it (tsk tsk), but a team from Santa Catarina State University in Brazil just might have the solution -- according to MIT Technology Review, they've cooked up an in-car hardware/software combo that can detect when you're on the phone and behind the wheel.
AI algorithm takes seat on investment company's board
The coming robot apocalypse will need robot executives to rule over it, but it looks like they're going to have to start with more mundane business first. A Hong Kong venture capital firm has just "appointed" an artificial intelligence tool called VITAL to its board of directors to help find promising investments. It'll scan things like financing, IP and clinical trials from prospective companies, share that info with the board and even cast the sixth vote. So far, it's helped find two promising outfits already, though it hasn't voted yet. Still, the main goal is to "draw attention to it as an independent decision maker," according to the company. And draw publicity to itself, no doubt.
Disney uses adorable little robots to illustrate big ideas
Disney Research has had some neat ideas in the past (capacitive touch feedback for plants, as an example), but the lab's amped up the cute factor lately. Its newest project? Getting tiny, LED-adorned robots to illustrate things such as The Big Bang. Like a good deal of Mickey's science projects, the experiment, dubbed "Pixelbots," is based around interactivity. The 2-inch swarm bots use magnetic wheels to move about on vertical surfaces. Meanwhile, an algorithm ensures that they won't hit one another and RGB diodes keep the robots looking pretty. Individual units can even be plucked out of formation and the pack will intelligently work to fix the gaps and reform the original shape.
BYU image algorithm can recognize objects without any human help
Even the smartest object recognition systems tend to require at least some human input to be effective, even if it's just to get the ball rolling. Not a new system from Brigham Young University, however. A team led by Dah-Jye Lee has built a genetic algorithm that decides which features are important all on its own. The code doesn't need to reset whenever it looks for a new object, and it's accurate to the point where it can reliably pick out subtle differences -- different varieties of fish, for instance. There's no word on just when we might see this algorithm reach the real world, but Lee believes that it could spot invasive species and manufacturing defects without requiring constant human oversight. Let's just hope it doesn't decide that we're the invasive species.
Quadrocopter drone recovers from failures without skipping a beat (video)
Quadrocopter drones are capable of some incredible acrobatics, but they seldom handle failure all that gracefully. ETH Zurich's Mark Mueller is tackling this problem through a new failsafe algorithm that gives these flying robots a better chance of survival. As you'll see in a video demo after the break, the software automatically compensates for rotor failures, bringing a drone back to its original position before giving the owner an opportunity to land the craft. Mueller's routine works even when there's just one propeller left, and it could eventually avoid dangerous objects on the way down. While there's no mention of when the algorithm will reach copters outside of the lab, there's a patent on the way -- we'd expect it to reach production drones at some point in the future.
Study finds app ratings now more critical to rankings than ever before
We knew Apple was tweaking its algorithms for ranking top apps, but a new report by app-testing firm Appurify gives us a much better idea of just how heavily weighted ratings are becoming. The company's study examines at the ratings of the top 1,000 apps on the App Store. Perhaps not surprisingly, the results show that more than half of the top 1,000 apps have a rating of at least 4.5 out of 5 stars, with 75% scoring 4 stars or higher. More interesting, however, is that the higher an app is placed in the charts, the more the ratings seem to matter. For example, in order to reach the top 300 of all apps in the marketplace, you're going to need hundreds of 4.5- and 5-star reviews every single day, but the criteria for landing in the top 300-600 apps is not nearly as demanding. Head on over to Appurify's blog to check out the full results. [via TechCrunch]
MIT algorithms teach robot arms to think outside of the box (video)
Although robots are getting better at adapting to the real world, they still tend to tackle challenges with a fixed set of alternatives that can quickly become impractical as objects (and more advanced robots) complicate the situation. Two MIT students, Jennifer Barry and Annie Holladay, have developed fresh algorithms that could help robot arms improvise. Barry's method tells the robot about an object's nature, focusing its attention on the most effective interactions -- sliding a plate until it's more easily picked up, for example. Holladay, meanwhile, turns collision detection on its head to funnel an object into place, such as balancing a delicate object with a free arm before setting that object down. Although the existing code for either approach currently requires plugging in existing data, their creators ultimately want more flexible code that determines qualities on the spot and reacts accordingly. Long-term development could nudge us closer to robots with truly general-purpose code -- a welcome relief from the one-track minds the machines often have today.
Magisto sharpens its AI video editing algorithm, adds themes, albums and group editing
Sure, Magisto's automatic video editing algorithms are great for social media mashup clips, but what if you want to use the service's robotic sense of cinema to tell a story? CEO Oren Boiman says it's just what social video is missing, and has tweaked the service to fill the gap. Users now have access to a collection of themes to change how their footage is handled. The idea is to tip the algorithm in on the emotion the user is trying to convey, selecting "so cute" or "street beat" to cue it to select appropriately adorable or aggressive song suggestions, special effects or title treatments. The service also added a new video album feature, making it easier to organize and share videos with friends and family, and hopes to implement a collaborative editing system soon -- complete with post-production tools to tweak the computer's direction. Of course, you could always do things the old fashioned way.
German robot arm learns ping-pong as it plays humans, might rival its masters
We like to tell ourselves that learning by doing is the best strategy for improving our skills, but we seldom apply that philosophy to our robots; with certain exceptions, they're just supposed to know what to do from the start. Researchers at the Technical University of Darmstadt disagree and have developed algorithms proving that robot arms just need practice, practice, practice to learn complex activities. After some literal hand-holding with a human to understand the basics of a ping-pong swing, a TUD robot can gradually abstract those motions and return the ball in situations beyond the initial example. The technique is effective enough that the test arm took a mere hour of practice to successfully bounce back 88 percent of shots and compete with a human. That's certainly better than most of us fared after our first game. If all goes well, the science could lead to robots of all kinds that need only a small foundation of code to accomplish a lot. Just hope that the inevitable struggle between humans and robots isn't settled with a ping-pong match... it might end badly.
Scientists develop pair of algorithms that could enable thermal cameras to pick out drunk people
We're not sure if Georgia Koukiou and Vassilis Anastassopoulos of the University of Patras in Greece like a tipple or not, but the pair have developed two algorithms that, when used with thermal imaging, could pick out drunk people in crowds. What is it that betrays your best intentions to look sober? As always, your face. Booze causes the blood-vessels in your visage to dilate, and the researchers used this principle to compare facial scans against a database of tipple-free mug shots. Likewise the duo found that when under the influence, the nose gets warmer, while the forehead cools -- another visual check that the infrared can help identify. The hope is that using this technology, law-enforcement can make a judgement call based on more than just your wonky walk. But in our experience, the troublemakers are pretty good at outing themselves.
Google's Turing doodle celebrates his genius, reminds us how dumb we are (video)
This week sees many corners of the globe celebrating the 100th anniversary of the birth of Alan Turing. A man whose contribution to the worlds of tech and gadgets is immeasurable -- a sentiment not lost on Google. Today, geeks and norms worldwide will be waking up to possibly the most complex doodle to date. Can you set the machine and spell out "Google"? If you can, you'll be sent off to lots more information about the man himself. This isn't the only thing Mountain View's done to keep his legacy alive, having previously helped Bletchley Park raise funds to purchase (and display) Turing's papers, and more recently helping curators at London's Science Museum with its Codebreaker - Alan Turing's Life and Legacy exhibition. If you haven't already, head to Google.com and pop your logic hat on, and if you get stuck, head past the break for a helpful video.
Netflix explains its recommendation system, can't find a reason for Adam Sandler's last movie
In case you've been wondering why Netflix tends to recommend the movies it does, there's a post on the company's Tech Blog breaking down the various levels of its system. Remember the Netflix Prize contest? Teams of researchers produced competing algorithms capable of more accurately predicting how members would rate movies, but while some of the early winning efforts are still in use, the million dollar solution was never implemented because the potential gains were too small to justify the engineering effort needed. Additionally, while Netflix still hasn't implemented individual profiles for household members yet, the blog indicates it does try to recommend something for everyone, seeking both accuracy and diversity -- which may explain some of more out there picks in our personal "recommended for you" list. Where available (read: outside the US) Facebook integration plays a part too, as well as a variety of information used to find movies similar to those previously viewed. The proof of how all these parts come together is ultimately judged by the viewers, so while we wait for part two of the post with more data to pore over -- is Netflix managing to accurately pull any flicks you want to watch out of its catalog?
Self-sculpting 'smart sand' can assume any shape, create instant prototypes (video)
A new algorithm developed by the Distributed Robotics Laboratory at MIT's Computer Science could lead to an exciting fast prototyping tool, being dubbed "smart sand." Immerse an object in the sand, tiny cubes that send simple proximity messages to each other, which relay through the swarm and determine which blocks are adjacent to the object to be modeled, and those that aren't. Using this data, it's possible to create a map of the subject to be replicated. Initial tests were performed using 2D models, but has also been shown to work reliably with 3D shapes also. While true smart sand would need "grains" much smaller than currently possible, it's said that this isn't an "insurmountable obstacle." The paper will be presented at the IEEE conference in May, or keep going past the break for the explanatory video.
Daily iPhone App: MyTunes Pro makes your music sound better
MyTunes has been on the App Store for a little while. It's developed by SRS that uses that company's patented sound enhancement algorithms to make music sound "better." It's not completely clear what the algorithm does (as SRS wants to keep it secret), but basically, the tech boosts and lowers certain qualities in audio to make it clearer and more listenable. MyTunes Pro is the new version of MyTunes. I saw it in action at CES earlier this year, and it's now available for the iPhone and the iPad (here's the HD version). The biggest update is that you can use AirPlay directly with the app. This lets you push your iTunes library through MyTunes Pro's enhancer, and then kick it out to an AirPlay system. You'll also find new controls for a system called "TruSpeed," which will speed up or slow down your audio without changing the pitch (so you can listen to more podcasts in less time). A new "workout mode" lets you gather songs by tempo (BPS). The interface of the app has been updated as well. Unfortunately, it's still pretty laggy (and doesn't really feel like a native interface), but it is a little easier to navigate and use than earlier versions. MyTunes Pro is free to try, so if you just want to hear what it does to your music, you can download it and give it a try (you get 10 minutes a day, unless you pay an in-app purchase of US$6.99). It's an app that's of somewhat limited use, because you can only listen to music in your iTunes library. iOS doesn't let audio from services like Pandora or Slacker get processed. Still, if you listen to music coming off of your iOS device all day, and would like it to sound even better (or just want to have a really capable EQ to use on it), MyTunes Pro should do exactly that.
