If you're heating your home you'd be paying that electric bill anyway. Whatever work is done in the meantime, ultimately the machine will convert pretty much 100% of the input watts into watts output as heat.
You left out the part where you prove "being black CIA" and "being DB Cooper" are independent. Only then the probability of being both is the product of the two probabilities.
Take a three-storied house, with eight windows on each floor. On the roof there are two dormer windows and two chimneys. On every floor there are two tenants. And now, tell me, gentlemen, in what year the valet's grandmother died.
#1. Post obvious fake news.
#2. See who retweets and shares it.
#3. Update DB of extremely gullible people.
#4. Focus subsequent propaganda better.
#5. Profit!
Unless the researchers have a steady need for the computations, moving work to the cloud makes more sense. Why build a multi-million dollar facility when you can just rent the computers for a day or two for your computation?
Of course we do. Top500 centres are almost exclusively used for scientific computations. In my field 128-256 GiB of RAM per node and an InfiniBand interconnect @40gbps+ are needed for reasonable performance, with typical jobs using hundreds of CPU cores and larger ones perhaps 8192 CPU cores. Most of sparse matrix algebra requires a lot of comms, things grind to a halt when the interconnect is slow. You can't do this kind of stuff in the cloud, not today.
So far, so good. But of course, Facebook did not go this way, i.e. no client side application to generate hashes.
In addressing your worries I was thinking in terms of how this could be solved, not in terms of how this is being handled currently.
Hmm. If I have the client side application that generates the hashes, I will be able to easily experiment with minor modifications, and can automate the process that turns minor changes into significant hash variations. Especially by defining areas of the original which must be preserved, and others which can be altered.
In transform space, I guess. In the bitmap these would be more like "intangible features" than areas. I agree in principle the hashes could be fooled, though.
And of course, by using wavelet transforms, you will make it really easy to automatically generate images that suppress valid images from your competitors, whose who say what you dislike, those who post incriminating information about your nefarious activities, etc.
Hell, if this is implemented, I will try to be the first to market a program to take advantage of the problems with your approach.
But that just returns to point five in your original statement ("how do you make sure that the mechanism is not used to suppress legitimate pictures?") which I did not take issue with and continue not to.
I'm not sure I understand how wavelets are going to help you with this. It's really not hard at all to modify an image such that the wavelet transform would be markedly different (heck, a brightening/darkening filter would do the trick, especially if you were only hashing the Y-Channel of an image). Unless you're using wavelets in a way I've never heard of before. One of my former jobs was developing image compression software so I have a reasonable working knowledge of how wavelet transforms work and I've personally implemented several variations of them.
I'm familiar with wavelets, but I don't have the kind of hands-on experience you say you have. I was basing my statement on the fact that a) google search by image seems to work OK, b) a slashdot article from maybe 2-3 years back which, AFAIR, presented quite a robust way of fingerprinting images that was (said to be) difficult to fool with "minor changes". This mentioned some kind of wavelet transform. I'm not vouching for how well it works in practice, maybe I was too optimistic in my assessment.
Slashdot Mirror
If you're heating your home you'd be paying that electric bill anyway. Whatever work is done in the meantime, ultimately the machine will convert pretty much 100% of the input watts into watts output as heat.
s/watts/joules/g
What kind of liquor goes best with antidepressants and chicken wings?
Diethyl ether. Pace yourself.
You can't possibly SHOUT at us CONFIDENTLY like you KNOW, and ask for an explanation simultaneously.
Are the numbers for heterocide similar?
it is a rare technology that has caused deaths in a fairly direct way.
Bill, this is gunpowder. Gunpowder, Bill.
False analogy much?
You left out the part where you prove "being black CIA" and "being DB Cooper" are independent. Only then the probability of being both is the product of the two probabilities.
Yo!
Take a three-storied house, with eight windows on each floor. On the roof there are two dormer windows and two chimneys. On every floor there are two tenants. And now, tell me, gentlemen, in what year the valet's grandmother died.
Climate is an average of the weather over a defined period. I.e. it is weather statistics. No weather event can be affected by statistics.
Just explained by it.
#1. Post obvious fake news.
#2. See who retweets and shares it.
#3. Update DB of extremely gullible people.
#4. Focus subsequent propaganda better.
#5. Profit!
Those small furry creatures, did they groove with a Pict by any chance?
Unless the researchers have a steady need for the computations, moving work to the cloud makes more sense. Why build a multi-million dollar facility when you can just rent the computers for a day or two for your computation?
Of course we do. Top500 centres are almost exclusively used for scientific computations. In my field 128-256 GiB of RAM per node and an InfiniBand interconnect @40gbps+ are needed for reasonable performance, with typical jobs using hundreds of CPU cores and larger ones perhaps 8192 CPU cores. Most of sparse matrix algebra requires a lot of comms, things grind to a halt when the interconnect is slow. You can't do this kind of stuff in the cloud, not today.
Human Mini-Brains Growing Inside Rat Bodies Are Starting To Integrate
Abbink and his team took a deep dive into IBM's archives. They were especially interested in the company's history in the postwar years,
I see what you did there.
So far, so good. But of course, Facebook did not go this way, i.e. no client side application to generate hashes.
In addressing your worries I was thinking in terms of how this could be solved, not in terms of how this is being handled currently.
Hmm. If I have the client side application that generates the hashes, I will be able to easily experiment with minor modifications, and can automate the process that turns minor changes into significant hash variations. Especially by defining areas of the original which must be preserved, and others which can be altered.
In transform space, I guess. In the bitmap these would be more like "intangible features" than areas. I agree in principle the hashes could be fooled, though.
And of course, by using wavelet transforms, you will make it really easy to automatically generate images that suppress valid images from your competitors, whose who say what you dislike, those who post incriminating information about your nefarious activities, etc.
Hell, if this is implemented, I will try to be the first to market a program to take advantage of the problems with your approach.
But that just returns to point five in your original statement ("how do you make sure that the mechanism is not used to suppress legitimate pictures?") which I did not take issue with and continue not to.
I'm not sure I understand how wavelets are going to help you with this. It's really not hard at all to modify an image such that the wavelet transform would be markedly different (heck, a brightening/darkening filter would do the trick, especially if you were only hashing the Y-Channel of an image). Unless you're using wavelets in a way I've never heard of before. One of my former jobs was developing image compression software so I have a reasonable working knowledge of how wavelet transforms work and I've personally implemented several variations of them.
I'm familiar with wavelets, but I don't have the kind of hands-on experience you say you have. I was basing my statement on the fact that a) google search by image seems to work OK, b) a slashdot article from maybe 2-3 years back which, AFAIR, presented quite a robust way of fingerprinting images that was (said to be) difficult to fool with "minor changes". This mentioned some kind of wavelet transform. I'm not vouching for how well it works in practice, maybe I was too optimistic in my assessment.
- how do you avoid charges of moving and storing child porn if the user is underage?
By computing the hashes client-side and transmitting and storing only the hashes, obviously.
- how do you make sure that minor changes to the original picture do not produce completely different signatures?
Wavelet transform. Compute hashes in wavelet space.
- how do you make sure that none of your employees have access to the originals?
By computing the hashes client-side and storing only the hashes, obviously.
How do we know it wasn't a weaponized asteroid intended to clearing and terraforming this planet for the new human species to evolve and be monitored?
Sent by the teapot, no less.
Say hi to the elves from me!
FB doesn't see her MAC address.
The lion cells don't know what kind of device they're in.
They're called cages, not cells.
Entanglement does not transfer information.
Vonnegut's "Player Piano", anyone?
Tech evangelist.