matsiyatzy

I've tried colorizing photos with this code (result here : https://twitter.com/matsiyatzy/status/706270684209680384) using EC2 instances. It's pretty fast, around 10-15 minutes, though you'll definitely want GPU-instances with CuDNN drivers installed (which can be a bit tedious to set up).

Here's a (currently incomplete) list of Deep Learning frameworks : https://docs.google.com/spreadsheets/d/1XvGfi3TxWm7kuQ0DUqYr...

There are many :)

You can download the file "synset_words.txt" with class numbers in caffe via running the script ./data/ilsvrc12/get_ilsvrc_aux.sh . But here's the file online as well : https://github.com/sh1r0/caffe-android-demo/blob/master/app/...

Probably because you've used up your "quota" of free economist articles. If you follow the link in incognito mode, the article is visible.

Also, link to the original paper : http://rsos.royalsocietypublishing.org/content/2/5/150081

Does anyone know if there is any code available for this paper? Would be extremely interesting to see comparisons of the EM-algorithm mentioned with regular SGD in terms of speed and precision.

Yep, the method will alert you when there is enough data to be able to conclude which hypothesis is true.

There's definitely geometric differences between male and female faces. The most distinct difference is that the male chin tends to be larger, giving the face a more "square" shape, while female faces have smaller chins, giving the face a slight "almond" shape. And I'm sure there's more subtle differences as well.

nice easter egg : following the "treasure hunt" leads to red letters spelling out "APRIL FOOLS".

Along each of the trails leading to the red letters, there are also hidden letters that spell out "MMC-900913" (a reference to last years google maps april fools)

All the letters together spell out "APRIL FOOLS"

It's a bit confusing that this has the same name as another (pretty well known) a/b test calculator: http://www.thumbtack.com/labs/abba/

Well, we've had similar drugs (though not vaccines) for a while I believe, and they don't always work as they should.

A guy I worked with was a recovering alcoholic on a "return-to-work" program. As a part of this program he had to take antabuse to stop him from drinking again. He died from a fatal heart attack within a month after he started working. It turns out he'd secretly still continued to drink even while on antabuse, which put an immense strain on his heart (antabuse gives tachycardia on consuming alcohol), causing the heart attack. Sad story.

Alcholics have a really strong urge to drink, even though it causes discomfort.

Hmm, that's interesting. Is there any name for that kind of method? I might look at implementing it.

Hi, creator of the headtrackr library here. As far as the headtrackr library goes, it does include a smoother (moving average) to avoid jittering of the tracking. It's set pretty low by default though, since heavy smoothing will lead to lag in tracking sudden movements. Any kind of moving average is a trade-off between responsiveness and smoothness, unfortunately...

Lighting is pretty important for the quality of the face tracking. If it doesn't work first time, try reinitializing a couple of times, the face detector tends to lock onto other objects than the face some times.

Check out http://auduno.github.com/headtrackr/examples/targets.html (though you'll need webgl support)

The point of the Surface "exercise" for Microsoft is not really to make a profit.

They're trying to "lead the herd" in creating a feasible merge of the "desktop" paradigm and the "mobile" paradigm, keeping Windows in the center. This is extremely important to Microsoft, since they have no foothold in the mobile market, and that's where the entire consumer market is headed. We can expect the merge of the desktop paradigm and mobile paradigm (i.e. keeping both battery-life, light weight and performance) to happen at some point in the future anyway, but meanwhile, Android and iOS/OS X are growing at a rapid pace, all at the cost of Windows marketshare. If mainstream users move to these OSs, the merge will happen with Android and iOS as center, and Microsoft will have a real rocky road ahead.

Trust me, Microsoft is willing to throw tons of money at this problem.

It should be mentioned that most of these things are already implemented by numeric.js (http://www.numericjs.com/)

Yeah, numeric.js is awesome and very fast compared to other libraries. Would be interesting to see a comparison in terms of performance between these two.

"Technology is anything that was invented after you were born.” - Alan Kay

I find this quote to be surprisingly true. It boggles the mind to think that mundane stuff such as sewing machines once was high technology, but there you go. My grandmother still remembers the marvel the day she got her first sewing machine.

For introverted or shy employees, the issue is not that it isn't easy to talk to the CEO/manager, it's just that they're unlikely to voice their opinions unless they're specifically asked about it. Which is what this meeting is about.

But yeah, for some employees it will be extremely useful, for others, not so much.

Well, it's usually very hard to copy other people's flash code that you might see somewhere. With webRTC/javascript, all you really need to do is "view source". This, plus mashups with other web APIs, makes webRTC a much more powerful platform than flash.

I would perhaps also have added a calculation of the power of the test. But I really love that they're actually explaining the details behind it, which a lot of other A/B-test calculators out there don't do.

Thanks! Interesting articles by the way.

I believe the model they're using is called Pareto/NBD and is fleshed out a bit more here:

http://marketing.wharton.upenn.edu/documents/research/Fader_...

I would expect a significant part of costs to be rent, as stores typically are in relatively expensive locations. And there's also considerable loss due to unsellable excess inventory, which is what the "fast fashion" business model tries to reduce.

I wasn't aware of that. That probably means the ideas behind this is a bit different, yes.

This is referred to as the "fast fashion" business model. H&M and Zara in Europe has been doing it for quite a while already. Wikipedia article here: http://en.wikipedia.org/wiki/Fast_fashion

It's not so surprising that they go for this business model, since "fast fashion" retailers have an average profit margin of 16%, compared to 7% for typical retailers, at least according to this source: http://www.google.com/url?sa=t&rct=j&q=&esrc=s&#...

Basically it's "just-in-time" production coupled with "agile"-type response with regards to fashion trends and customer wants.

While it's true that Sony has messed up their opportunities with DRM and the like, I highly doubt that Sony realistically could do what Apple did, simply because competing record labels would never in their right mind license their music in this way to another competitor. iTunes on the other hand, was just another distribution channel to the record labels, and not a direct competitor.