haraldurt

The percentages aren't quite additive like that. Say Epic gets $1 per sale, and the consumer pays $1.3 with the 30% Apple cut. A discount of ~23% (1/1.3) gets the sale price back down to $1. Include chargebacks and such for the missing ~2%.

Is it preferred to release two bad actors or to release two good actors? Being in prison is a punishment, but it is also rehabilitation (ideally).

There have been some fascinating steps taken towards that direction, e.g. [0] and its related work. There they take the approach of transferring input images back and forth domains (think, a street imaged in summer transferred to its winter manifestation and back or a synthetic GTA image to real-world and back are the examples in [0]). Doing this while simultaneously holding the semantic content of the input unchanged with a GAN-type strategy seems to be a way to coerce the neural net's internal representations to capture what we want them to instead of idiosyncrasies of the dataset.

[0] https://arxiv.org/abs/1711.03213

Augmenting your training dataset with adversarial examples is known as adversarial training, see e.g. [0] for a recent overview with empirical results. This seems to be a good first step in defending against such attacks, though the most naive approach of adversarial training doesn't work as well as you'd expect.

[0] https://openreview.net/forum?id=rkZvSe-RZ

The first plot of "Activation 0" appears to actually be the random input, if it corresponds to hs[0] in the code. The rest of the activation plots seem to all be strictly non-negative. The other plots with negative values are of gradients, not activations.

Per Wikipedia, Hans Jonatan worked in Djúpivogur and lived in Borgargarður (which I believe is/was a farm or homestead located in the former) in the East Fjords of Iceland.

https://en.wikipedia.org/wiki/Hans_Jonatan#Life_in_Iceland

Not necessarily. Adversarial examples have been shown to, for instance, be transferable across different networks with different hyperparameters (e.g., number of layers) trained on disjoint subsets of a training set [0, section 4.2]. There are more references from the paper linked by the OP.

[0] https://arxiv.org/abs/1312.6199

In one sense, no. You can guarantee privacy of any given input (or any subset of k inputs) by applying transfer learning of an ensemble of models trained on subsets of the training data [0][1]. This is useful if, for instance, you train on medical data and you don't want anyone to know that "John Doe, HIV+" was part of the input. If your adversary does not take such precautions, however, then your canary should work.

[0] https://arxiv.org/abs/1610.05755

[1] https://static.googleusercontent.com/media/research.google.c...

This exchange also included in the linked New Yorker post:

>Over the decades, Dylan and Cohen saw each other from time to time. In the early eighties, Cohen went to see Dylan perform in Paris, and the next morning in a café they talked about their latest work. Dylan was especially interested in “Hallelujah.” Even before three hundred other performers made “Hallelujah” famous with their cover versions, long before the song was included on the soundtrack for “Shrek” and as a staple on “American Idol,” Dylan recognized the beauty of its marriage of the sacred and the profane. He asked Cohen how long it took him to write.

> “Two years,” Cohen lied.

> Actually, “Hallelujah” had taken him five years. He drafted dozens of verses and then it was years more before he settled on a final version. In several writing sessions, he found himself in his underwear, banging his head against a hotel-room floor.

> Cohen told Dylan, “I really like ‘I and I,’ ” a song that appeared on Dylan’s album “Infidels.” “How long did it take you to write that?”

> “About fifteen minutes,” Dylan said.

> When I asked Cohen about that exchange, he said, “That’s just the way the cards are dealt.” As for Dylan’s comment that Cohen’s songs at the time were “like prayers,” Cohen seemed dismissive of any attempt to plumb the mysteries of creation.

Iceland became independent in 1945[0]; in 1918 Iceland became a sovereign state under the rulership of the Danish king[1].

The Independence Party is center-right (very conservative for Iceland) and was formed in 1929 advocating for full independence of Iceland. They haven't changed their name since. It's been one of the four major political parties of Iceland since inception and more often then not the largest; every head of the party has been prime minister at some point (except the current one; that might change in the next week as they are in coalition with the just-resigned prime ministers party and there is a void to fill). In short you could say that they are pro-business and for smaller government. The overview here [2] is a good start, but you can go straight to the source here (english below) [3].

[0] https://en.wikipedia.org/wiki/Founding_of_the_Republic_of_Ic...

[1] https://en.wikipedia.org/wiki/Danish%E2%80%93Icelandic_Act_o...

[2] https://en.wikipedia.org/wiki/Independence_Party_%28Iceland%...

[3] http://www.xd.is/um-sjalfstaedisflokkinn/

There's a start up named Affectiva which does exactly that. Emotion measurement from video which is used for, among other things, targeting ads and marketing campaigns.

https://en.wikipedia.org/wiki/Affectiva

>The average distance between any two of Facebook’s 1.5 billion users is 4.74 edges. Sorry Kevin. With 1.5 billion users the whole graph can be explored within 5 hops.

Nitpicking here, but that does not follow. Starting from a random user and hopping 5 times in all directions you'd expect to explore about half of the graph.