Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

iamtrask.github.io

Post reply on HN
Print view View original

41–50 of 54 posts

Re: Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

#41

Post by rocqua »

Ok, I'm probably missing something.

How is public-key homomorphic encryption possible? Suppose I know the pub-key, and have some cipher text X. Can't I simply encrypt 0 and try, for all plausible values V, whether encrypt(V) - X = encrypt(0). Or is the encryption function not reversible, i.e. there are multiple 0s?

If not, it seems like you'd need quite a bit of true entropy in the plausible values. I don't see how you add something like a Nonce to artificially add entropy.

Heck, if you have nice integer under/over flow on division, that would give a crude implementation of comparison, bringing the search down by a logarithm.

It seems like the only reasonable situation is one where indeed multiple different values decrypt to 0.

Re: Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

#43

Post by murbard2 »

rocqua wrote: post #41

Ok, I'm probably missing something. How is public-key homomorphic encryption possible? Suppose I know the pub-key, and have some cipher text X. Can't I simply encrypt 0 and try, for all plausible values V, whether encrypt(V) - X = encrypt(0). Or is the encryption function not reversible, i.e. there are multiple 0s? If not, it seems like you'd need quite a bit of true entropy in the plausible values. I don't see how y…

> Or is the encryption function not reversible, i.e. there are multiple 0s?

Yep, there's padding.

Re: Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

#44

Post by zo1 »

petre wrote: post #20
TTPrograms wrote: post #18

The number of people killed by terrorism in developed countries per year is, what, a hundredth of those killed in car accidents? A tenth of those shot in Chicago alone? http://www.businessinsider.com/death-risk-statistics-terrori... Terrorists have to kill literally hundreds of times more people per year for it to justify even the meagerest response. And one may argue about the economic impacts, but that's just a ref…

If the media wouldn't cover terrorism it would probably go away by itself. Israel is probably the only country dealing with it effectively.

That's like telling a bullied kid to ignore a bully (or stop giving him/her attention) and that it will make the bully go away, eventually, sometime in the future (maybe). It may work in the long-term, but it doesn't fix the problem of people hurting and being scared in the now.

I'm a simple person. We have government, and we pay taxes so we can get services and have safety (that is what they tell us). So, terrorism is something we expect government to fix and resolve. Just like all the other things listed above that kill more people per year than terrorism. We can't just say "let's ignore it", and "we have to deal with X acts of terrorism per year for a N years before the terrorists lose interest". That is what some of these "anti-anti-terrorism" arguments logically lead to, and rightfully it's a tough pill to sell to the public.

Re: Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

#45

Post by bilotrace »

Several problems with your points - Data

For the SPAM example you provided, you used a Data Set available publicly with no consequences. But where will the training data for prediction of homicides will come from? Will it be accurate? Will it unfairly target minorities?

The predictor uses features in the data to predict the future. What features will the machine learning algorithm learn to predict homicide? location? age? gender? ethnicity? mannerism? income? I could see big problems with any features used to detect and potentially send law enforcement for "Investigations"

- Detectors

Capturing and sending users data to a warehouse is a privacy/security risk. But what is worse is installing a detector in their house, or in their computer. What happens when a citizen removes the SPAM detector from their computer? Well, the next logical step is to pass a law requiring all citizens to run a specific process on their computer and are not allowed to reverse engineer it or else...

For me, this is even worse than having my privacy violated. It would mean users would not have root access to their own computer. It would mean, many applications will be illegal and developers cannot write certain applications that enable people to send emails. If this is not impossible, it would be the worst outcome for everyone.

- Globalization

The SPAM detector, the fire detector and the sniffer dog are great localized examples. But todays problems are globalized. Attacks might be planned and coordinated from different country with different set of rules. And not all countries are considerate when it comes to privacy. How will global security surveillance deployment work when we can't even agree on matters of climate change?

Re: Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

#46

Post by aub3bhat »

ja30278 wrote: post #32
TTPrograms wrote: post #18

The number of people killed by terrorism in developed countries per year is, what, a hundredth of those killed in car accidents? A tenth of those shot in Chicago alone? http://www.businessinsider.com/death-risk-statistics-terrori... Terrorists have to kill literally hundreds of times more people per year for it to justify even the meagerest response. And one may argue about the economic impacts, but that's just a ref…

Counterpoint: As Taleb points out, car accident deaths are 'thin-tailed'; that is, the rate of car fatalities is essentially fixed (and predictable by looking at past data), while terrorism is 'fat-tailed', and the number of deaths that would be caused by a dirty-bomb in Manhattan does not appear in historical data, and so it can't be 'priced' in the same way.

I have made this exact argument (quoting Taleb) but HN users are not receptive to anything other than a do-nothing approach.

https://news.ycombinator.com/item?id=14262011

Re: Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

#47

Post by rocqua »

murbard2 wrote: post #43
rocqua wrote: post #41

Ok, I'm probably missing something. How is public-key homomorphic encryption possible? Suppose I know the pub-key, and have some cipher text X. Can't I simply encrypt 0 and try, for all plausible values V, whether encrypt(V) - X = encrypt(0). Or is the encryption function not reversible, i.e. there are multiple 0s? If not, it seems like you'd need quite a bit of true entropy in the plausible values. I don't see how y…

> Or is the encryption function not reversible, i.e. there are multiple 0s? Yep, there's padding.

Whats to prevent me from finding 'almost all' zeros by trying enough (say 10 000 000) versions of enc(v) - enc(v).

That, combined with comparison based on integer division and underflow would still make decryption quite easy.

Re: Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

#48

Post by murbard2 »

rocqua wrote: post #47
murbard2 wrote: post #43

Earlier quoted context omitted.

> Or is the encryption function not reversible, i.e. there are multiple 0s? Yep, there's padding.

Whats to prevent me from finding 'almost all' zeros by trying enough (say 10 000 000) versions of enc(v) - enc(v). That, combined with comparison based on integer division and underflow would still make decryption quite easy.

replace 10 000 000 with 2^256

Re: Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

#49

Post by mmirate »

__jal wrote: post #25
petre wrote: post #20

Earlier quoted context omitted.

If the media wouldn't cover terrorism it would probably go away by itself. Israel is probably the only country dealing with it effectively.

Not just the media. A lot of current U.S. politicians apparently believe their best path to power is endless fear mongering. And it works pretty well.

That's interesting. While studying the Reichstag Fire in grade-school history class, I ran across a quote from someone in the Nazi propaganda machine, who asserted something to the effect that exerting control over a democracy was simply a matter of manufacturing a large enough exterior threat for them to fear. Sadly, I cannot remember the quote nor the author.

Re: Safe Crime Prediction: Encrypted Deep Learning for Less Intrusive Surveillance

#50

Post by __jal »

mmirate wrote: post #49
__jal wrote: post #25

Earlier quoted context omitted.

Not just the media. A lot of current U.S. politicians apparently believe their best path to power is endless fear mongering. And it works pretty well.

That's interesting. While studying the Reichstag Fire in grade-school history class, I ran across a quote from someone in the Nazi propaganda machine, who asserted something to the effect that exerting control over a democracy was simply a matter of manufacturing a large enough exterior threat for them to fear. Sadly, I cannot remember the quote nor the author.

It was a well-discussed idea at the time, so I'm not sure who you're referring to. But Hermann Goering famously discussed it with Gustave Gilbert:

"[...] the people can always be brought to the bidding of the leaders. That is easy. All you have to do is tell them they are being attacked and denounce the pacifists for lack of patriotism and exposing the country to danger. It works the same way in any country."

Post reply on HN