Become a fan of Slashdot on Facebook

 



Forgot your password?
typodupeerror
Note: You can take 10% off all Slashdot Deals with coupon code "slashdot10off." ×

Comment Re:Do they actually work well now? (Score 5, Informative) 45

Last time I looked there was no application of ANNs which couldn't be solved more efficiently by other algorithms ... and the best ANNs used spiking neurons with Hebbian learning which are not amenable to efficient digital implementation.

Is it possible that last time you checked was a long time ago? Deep neural networks are again all the rage now (i.e. huge teams working with them at Facebook and Google) because

  1. (1) They have resulted in a significant performance improvement over previously state-of-the-art algorithms in many application tasks,
  2. (2) Although they are computation-heavy, they are amenable to massive parallelization (modern computational power is probably the main reason why they have improved singificantly with respect to ANNs of the 80-90s, given that the main architecture itself has not changed a lot, except possibly for the "convolution" trick which effectively introduces hard-coded localization and spatial invariance).

Check the wikipedia page for "convolutional neural networks" as well as other /. entries: http://slashdot.org/tag/deeple... , and from yesterday http://tech.slashdot.org/story... .

AI

Facebook AI Director Discusses Deep Learning, Hype, and the Singularity 71

An anonymous reader writes In a wide-ranging interview with IEEE Spectrum, Yann LeCun talks about his work at the Facebook AI Research group and the applications and limitations of deep learning and other AI techniques. He also talks about hype, 'cargo cult science', and what he dislikes about the Singularity movement. The discussion also includes brain-inspired processors, supervised vs. unsupervised learning, humanism, morality, and strange airplanes.

Comment Re:Overly broad? (Score 1) 422

I hold a faculty position in statistics (that's for the AC above who called me a "passer-by sitting at home in their boxers munching on Hot Pockets", so I guess I have to pull credentials, though in his defense my post sounded more dismissive than what I'd wanted).

Yes, the p-value threshold of 0.05 is considered "standard" in many applied sciences, in particular medicine. It is convenient for many of reasons that were outlined by other posters (cost, number or persons required for an experiment, ethics). It does not mean that it is intellectually satisfactory. The joke among statisticians is that this value was introduced about 100 years ago by the R.A. Fisher (one of the founding fathers of statistics) who once wrote something akin to "if we decide on a value of alpha such that the probability of falsely claiming a discovery when the null hypothesis holds seems reasonably low, say for instance, alpha=5%...", and this has somehow been engraved as gospel ever since.

The truth is, this threshold value of 5% is now considered very lax by modern statisticians, essentially because of the very large numbers of published papers reporting significant values as compared to Fisher's times. The posts of penguinoid and ras above explained it very professionally, one can also refer to "Why Most Published Research Findings Are False" (Note: this was published in PLOS medicine, hardly an obscure journal)

In conclusion, my post was certainly not a defense of soda pop (there is already sufficient evidence that it is extremely damaging for your health for very clearly identified reasons), but a reminder that the specific results of this study (the effect on telomeres), though certainly not to be dismissed, should not be considered as established truth at this point, but rather pointing in a direction which should be investigated further for confirmation. That, by the way, is the actual meaning of "being skeptical", unfortunately this tends to be conflated with "being in obtuse denial" nowadays.

Comment Re:Overly broad? (Score 5, Insightful) 422

A p-value of 0.04? This is a rather meager statistical significance. Mark me skeptical until the study has been reproduced independently.

For all I know they might have been looking at a lot of different nutrition factors and only reported those which appeared significant after the experiment (obligatory xkcd reference: http://xkcd.com/882/ )

Comment Re:Slashdot comments indicative of the problem (Score 1) 1262

I think everyone but the media recognizes at this point that the Quinn scandal is about corruption in journalism.

Except it's not. If it was, where is the outrage on an even remotely comparable scale concerning the truly powerful forces of corruption in this industry -- financial pressure being exerted by publishers on gaming outlets, flow of free perks being offered to journalists, secretly sponsored Let's Play videos, etc? Even if it was actually true that Quinn's admittedly particularly shitty personal behavior was only fueled by the desire of personal gain and media exposure (for which the evidence is nonexistent if you ask me), how would that even compare in terms of leverage gained and scale to the corrupting power of money, which is pervasive in this industry?

It's not about the corruption. It's about the sex, it's about the hate of anything that says "feminism", it's about the desperate quest to find a negative poster child justifying that hate towards anyone else expressing a related opinion. Using the fight against corruption as a justification is a total delusion, yet one in which all the haters have to believe, for otherwise they could not stand to face their own cognitive dissonance.

Comment Re:sensationalism? (Score 1) 196

You can solve this using Excel, but a dedicated app to to track the scenario mentioned in the original piece could be very useful to some.

As a matter of fact, it already exists: http://www.kittysplit.com/ This is a free webapp developed by some people I know. Also, probably prior art or something.

Comment Strategic analytic thinking for 4rth graders? (Score 1) 488

So only 1/3 of fourth graders were able to find the experimental setup to find the best fertilizer level out of nine, when you are only allowed to try six out of them.

The correct strategy consists in going in two steps, first trying out interspaced levels e.g. 2-4-6-8 then "refining" with the two remaining tries around the approximate minimum. This necessitates to model implicitly/intuitively the plant growth as a unimodal (increasing, then decreasing) function of the fertilizer level, thinking ahead with the limited tries constraint, and mentally planning different outcomes of the two steps.

I'd go contrary to the flow and say that 33% of 4rth graders solving an assignment of this difficulty is pretty darn awesome.

Comment Re:Happened to Me (Score 1) 107

There is a difference between randomly sprinkling a paper with references in a superficial effort to make it look "serious" and conform to the usual academic mold; and actively researching, citing and discussing earlier relevant references in comparison to your own work in a balanced way. The latter is how good quality academic writing should be done. The former tends to give rise to papers with pointless laundry lists of citations. I hope your friends were suggesting the latter way. Even if they were not able to point to specific references because they are not specialists of the issue you are addressing, they probably know by experience that it is quite unusual that no previous relevant references exist on a given academic issue. The fact is, nothing annoys a reviewer or an editor more than someone reinventing the wheel and giving the impression of ignoring previous work out of intellectual laziness.

Where there is a clear problem is when an editor or reviewers imposes an obviously irrelevant citation for self-serving reasons.

Image

Officials Say "Capes For the Unemployed" Plan Not Super Screenshot-sm 392

After what must have been an epic marketing meeting, a Florida unemployment agency decided to give 6,000 red capes to the jobless as part of its "Cape-A-Bility Challenge" public relations campaign. The capes cost $14,000 (not a bad price for 6k capes actually) and featured a cartoon character named "Dr. Evil Unemployment." As one might imagine, officials are calling for an investigation to be launched. It's a good thing there are an abundance of caped do-gooders without jobs in the area who should be able to help.
Google

Google Tests Multiple Account Login 122

tekgoblin noted joyous rumors for anyone forced to use multiple Google accounts "Wouldn't it be great if you could log into all of your Google accounts at the same time if you have multiple? Well it seems that Google may be implementing a way to do this in the near future. Right now it can be done with scripts such as a Greasemonkey script, but that isn't as easy as Google doing it for us. The people over at Google Operating System have had users submit a screenshot of what looks like a beta test for multiple account login. It appears that it will be available for Calendar, Code, Docs, Gmail, Reader, and Sites for the test but surely it will be across all Google apps when it's released."
Space

Ancient Comet Fragments Found In Antarctic Snow 92

An anonymous reader writes with this excerpt from Cosmos Magazine: "Two tiny meteorites recently recovered from Antarctic snow contain material dating back to the birth of our Solar System, and may provide clues about the delivery of organic matter to Earth. Researchers believe that these micrometeorites likely came from the cold, comet-forming outer regions of the gas and dust cloud that comprised the early Solar System, and sample its composition. Discovered in 2006, the particles measure less than 0.25 mm across and survived their journey through Earth's atmosphere relatively unscathed. More importantly, scientists found that they contain unusually high amounts of organic matter."
The Internet

Anyone Can Play Big Brother With BitTorrent 436

An anonymous reader writes "I was at the 3rd USENIX Workshop on Large-Scale Exploits and Emergent Threats yesterday, and there were people from the French Institute for Computer Science who have continuously spied on most BitTorrent users on the Internet for 100 days, from a single machine. They've also identified 70% of all content providers; yes, those guys that insert the new contents into BitTorrent. As a BitTorrent user, I was shocked that anyone with a box connected to the Internet can spy on what everyone is downloading on BitTorrent."
Patents

Is the Tide Turning On Patents? 172

Glyn Moody writes "The FSF has funded a new video, 'Patent Absurdity: how software patents broke the system,' freely available (of course) in Ogg Theora format (what else?). It comes at a time when a lot is happening in the world of patents. Recent work from leading academics has called into question their basis: 'The work in this paper, and that of many others, suggests that this traditionally-struck "devil's bargain" may not be beneficial.' We recently discussed how a judge struck down Myriad Genetics's patents on two genes because they involved a law of Nature, and were thus 'improperly granted.' Meanwhile, the imminent Supreme Court ruling In re Bilski is widely expected to have negative knock-on effects for business method and software patents. Is the tide beginning to turn?"
Encryption

Quantum Encryption Implementation Broken 133

I Don't Believe in Imaginary Property writes "Professor Johannes Skaar's Quantum Hacking group at NTNU have found a new way to break quantum encryption. Even though quantum encryption is theoretically perfect, real hardware isn't, and they exploit these flaws. Their technique relies on a particular way of blinding the single photon detectors so that they're able to perform an intercept-resend attack and get a copy of the secret key without giving away the fact that someone is listening. This attack is not merely theoretical, either. They have built an eavesdropping device and successfully attacked their own quantum encryption hardware. More details can be found in their conference presentation."

"Just think, with VLSI we can have 100 ENIACS on a chip!" -- Alan Perlis

Working...