It's not less secure, it's essentially useless. I'm afraid that an anagram would not make a good encryption scheme and be rather trivial to break. As in I could do it by hand.
The entire point of encryption is for the relationship between input and output to be highly chaotic. If you encrypt values that are similar, the values you get are not similar. In the same way, if you decrypt values that are similar, the values you get are not similar.
Speaking in terms of numbers: if you encrypted the numbers 5, 6, 7, then the three outputs should usually not be 21, 20, 22 (because they're also close together, which would make the encryption easily broken). If your range is 0-255, you should get something like 111, 75, 121.
Lossy compression and then decompression will change your numbers to something similar. Lets say you compressed 111, 75, 121, and a bunch of other numbers. When you decompress, you get 115, 76, 121 and a bunch of other numbers.
Because of the chaotic relationship between input and output in decryption, this decrypts to, say, 85, 118 and 136 instead of 5, 6 and 7. Because these aren't close to the original values, you can no longer make sense of the data.
The only way that a secure encryption algorithm could be used over lossy compression is (as Icebike said) if you could somehow force the compression algorithm not to compress the encrypted data. Whether that will be possible depends entirely on the compression algorithm.
Her husband acquired it in good faith; which only means that he got it from someone or somewhere and did not believe it to be stolen.
However, NASA owns it and it she therefore did not have the RIGHT to sell it, even if it would not have been a CRIMINAL OFFENSE for her to do so. Unless she knew that she doesn't own it, which she does now.
I agree with your sentiment, but I would like to correct you here, even at this late time, because you touch on a very important point.
The limbs are not "non-vital areas". There are major arteries in the arms and legs and a bullet in a limb is often a lethal wound if not treated immediately. Please don't spread Hollywood biology.
If you have means I highly recommend it. The first few hours after anesthesia is like Disney Land, but much cooler.
It is not just about being able to afford it. I have had general anaesthesia twice and it was horrible.
First, I puke all day Second, I seem to go on a bad trip. Seem to, since I can not remember any of it. But then I have to endure tales of everything I did, like kicking the doctor (both times).
I can't mod you up, so I'll just say that yours is the most true comment on the story.
It is also a fact that negative reinforcement is ineffective if not applied quickly. Locking someone up over a period of years is less than useless, but a "caning" might actually achieve something.
And if someone is truly so dangerous that we can not let him out again, is execution not less cruel?
I think youre the only other person on earth I have run into that knows about this game.
Really? My friends and I used to love it.
If you don't know about it, you might want to check out UFO:AI (ufoai.ninex.info, or find it on sourceforge). It's very playable but gets boring late in the game. That should improve as the game matures.
Some support circus administered the computers. A friend of mine looked over the guy's shoulder once, and I didn't believe him until he demonstrated that it works.
You're about a billion steps from showing equivalence between those two things.
I said that already:
Because of this, we can't say exactly how they work, and I would therefore be hard-pressed to give a formal proof of universality
I have already given you proof of the computational universality of the brain. (But the proof does not involve NNs, which would be the more interesting proof because the long-term goal would be to build wetware computers with a constructive proof.)
They approximate similar patterns, but are in no way the same thing.
Biological NNs work on the same basic principles as artificial NNs. Also, the definition of neural network is rather broad. Let me grab a book:
Definition: Neural computing is the study of networks of adaptable nodes which, through a process of learning from task examples, store experiential knowledge and make it available for use.
This definition is such that the neural nets of the living brain are included in the field of study.
(Aleksander, I. & Morton, H. An introduction to neural computing. (p. 1))
You can't go from Neural Network is Turing Complete to human brain is a Neural Network, therefore the human brain is Turing Complete.
So we know that the brain is computationally universal, and we know that some neural networks are computationally universal, and we know that the brain is a neural network. If this isn't clear evidence that the computational universality of neural networks in general extends to the brain, then please enlighten me.
It's not been done yet, and we don't understand enough to even think about such a thing.
Why do you say so?
The computational universality of a number of artificial neural network designs has been proven. Of course, that doesn't make a specific network necessarily universal.
Turing's machine assumed that you could have as much tape as you needed. The equivalent in feed forward neural networks (universal, and apparently common enough in the brain) is to assume that you can have as many neurons in the hidden layer (only one required for universality) as needed and that the order (highest number of inputs to a single neuron) is sufficiently high. This assumption is waived in practice, because otherwise I wouldn't be typing this on a "computer".
Biological neural networks are far more complex than artificial ones. Because of this, we can't say exactly how they work, and I would therefore be hard-pressed to give a formal proof of universality, but even the simplest artificial NN models (e.g. the McCulloch-Pitts model) are universal.
One more minor thing (in agreement with what you said): there are some indications in the literature that there may be some neural network models that are greater than a Turing machine. This is subject to a lot of very shaky conditions. I'm not up to date with the latest on this front; but if someone claims the brain is more than a Turing machine, he'd better show some proof.
tl;dr answer: Here's a simple proof: If I describe a specific Turing machine to you, you can calculate its result, assuming you have enough paper or a big and complex enough brain.
to forge ahead all at once without trying to understand every step and the implications of every signal
That is, in fact, exactly what is being done.
Neural networks (both biological and artificial ones) "overparameterise", which basically means that they carry redundant information.
Due to this overparameterisation, it is mathematically impossible to assign an exact meaning to each connection weight*. Hence, it is impossible to assign exact meaning to a given signal coming into a neuron in the general case. You must consider multiple (often all) connections simultaneously.
All it would take is understanding of how to keep these cells alive and connecting
Incorrect. Neural network design is complicated. Not every network topology is suited to solving every kind of problem. Whether you use a threshold function or a continuous transfer function, and exactly which functions you use, impacts what problems you can solve and how well***. Granted, some designs are computationally universal, but this doesn't guarantee that the network will be easy to train, or that you won't need a trillion neurons to solve a given problem with a theoretically suitable design.
Furthermore, if the initial connection weights* are chosen poorly, the network may take a very long time to train or it may train towards a garbage solution.
You have heard that NNs are self-organising, but this is only true within certain limits. Neural networks are no panacea. Each network can approximate the solution to a given computational problem (or find the exact solution), assuming that there are enough neurons, assuming furthermore that the initial design was good, and also assuming that the training method is reasonable**. The brain is a collection of many different networks, of different topology and with many very different kinds of cells, that interact.
IANA neurobiologist, but IAA grad student studying artificial NNs.
* Biological neurons' inputs are a little more complicated than simply assigning a weight to each input and summing them, but the analogy holds. (Dendritic computation is still little understood.)
** For example, if you want a single hidden-layer feed forward NN to approximate a curve and training data is randomly distributed around the curve with a skewed distribution, the LMS algorithm is not ideal. (It may still be good enough, depending on specifics.)
*** Here I speak of artificial NNs. There are analogous differences in biological ones, such as the type of synapse (chemical/electric, axon-dentrite/axon-axon/dendrite-dendrite/axon-soma, unidirectional/bidirectional) and dentritic computation.
Slashdot Mirror
No-one can guess my Slashdot password!
It's not less secure, it's essentially useless. I'm afraid that an anagram would not make a good encryption scheme and be rather trivial to break. As in I could do it by hand.
The entire point of encryption is for the relationship between input and output to be highly chaotic. If you encrypt values that are similar, the values you get are not similar. In the same way, if you decrypt values that are similar, the values you get are not similar.
Speaking in terms of numbers: if you encrypted the numbers 5, 6, 7, then the three outputs should usually not be 21, 20, 22 (because they're also close together, which would make the encryption easily broken). If your range is 0-255, you should get something like 111, 75, 121.
Lossy compression and then decompression will change your numbers to something similar. Lets say you compressed 111, 75, 121, and a bunch of other numbers. When you decompress, you get 115, 76, 121 and a bunch of other numbers.
Because of the chaotic relationship between input and output in decryption, this decrypts to, say, 85, 118 and 136 instead of 5, 6 and 7. Because these aren't close to the original values, you can no longer make sense of the data.
The only way that a secure encryption algorithm could be used over lossy compression is (as Icebike said) if you could somehow force the compression algorithm not to compress the encrypted data. Whether that will be possible depends entirely on the compression algorithm.
I think Dwedit was joking.
Lossy compression of encrypted data would mean little to no data. The other way around works.
The AC speaks truth! (Well I didn't let it finish loading, but the browser was connecting to goatse.ru...)
You're right. It sounded a bit overenthusiastic.
No hard feelings towards anyone.
Well, I DID RTFA.
Her husband acquired it in good faith; which only means that he got it from someone or somewhere and did not believe it to be stolen.
However, NASA owns it and it she therefore did not have the RIGHT to sell it, even if it would not have been a CRIMINAL OFFENSE for her to do so. Unless she knew that she doesn't own it, which she does now.
hands, feet, arms, legs
I agree with your sentiment, but I would like to correct you here, even at this late time, because you touch on a very important point.
The limbs are not "non-vital areas". There are major arteries in the arms and legs and a bullet in a limb is often a lethal wound if not treated immediately. Please don't spread Hollywood biology.
And people ask me why I don't want to be a programmer.
If you have means I highly recommend it. The first few hours after anesthesia is like Disney Land, but much cooler.
It is not just about being able to afford it. I have had general anaesthesia twice and it was horrible. First, I puke all day Second, I seem to go on a bad trip. Seem to, since I can not remember any of it. But then I have to endure tales of everything I did, like kicking the doctor (both times).
I can't mod you up, so I'll just say that yours is the most true comment on the story.
It is also a fact that negative reinforcement is ineffective if not applied quickly. Locking someone up over a period of years is less than useless, but a "caning" might actually achieve something.
And if someone is truly so dangerous that we can not let him out again, is execution not less cruel?
I find our society disturbing.
I think youre the only other person on earth I have run into that knows about this game.
Really? My friends and I used to love it.
If you don't know about it, you might want to check out UFO:AI (ufoai.ninex.info, or find it on sourceforge). It's very playable but gets boring late in the game. That should improve as the game matures.
Yes.
Some support circus administered the computers. A friend of mine looked over the guy's shoulder once, and I didn't believe him until he demonstrated that it works.
ah geez. it's like being back in school. my best mate's password was "123".
Ah, the memories. (The school's admin password was "access".)
You're about a billion steps from showing equivalence between those two things.
I said that already:
Because of this, we can't say exactly how they work, and I would therefore be hard-pressed to give a formal proof of universality
I have already given you proof of the computational universality of the brain. (But the proof does not involve NNs, which would be the more interesting proof because the long-term goal would be to build wetware computers with a constructive proof.)
They approximate similar patterns, but are in no way the same thing.
Biological NNs work on the same basic principles as artificial NNs. Also, the definition of neural network is rather broad. Let me grab a book:
Definition: Neural computing is the study of networks of adaptable nodes which, through a process of learning from task examples, store experiential knowledge and make it available for use.
This definition is such that the neural nets of the living brain are included in the field of study.
(Aleksander, I. & Morton, H. An introduction to neural computing. (p. 1))
You can't go from Neural Network is Turing Complete to human brain is a Neural Network, therefore the human brain is Turing Complete.
So we know that the brain is computationally universal, and we know that some neural networks are computationally universal, and we know that the brain is a neural network. If this isn't clear evidence that the computational universality of neural networks in general extends to the brain, then please enlighten me.
It's not been done yet, and we don't understand enough to even think about such a thing.
Why do you say so?
The computational universality of a number of artificial neural network designs has been proven. Of course, that doesn't make a specific network necessarily universal.
Turing's machine assumed that you could have as much tape as you needed. The equivalent in feed forward neural networks (universal, and apparently common enough in the brain) is to assume that you can have as many neurons in the hidden layer (only one required for universality) as needed and that the order (highest number of inputs to a single neuron) is sufficiently high. This assumption is waived in practice, because otherwise I wouldn't be typing this on a "computer".
Biological neural networks are far more complex than artificial ones. Because of this, we can't say exactly how they work, and I would therefore be hard-pressed to give a formal proof of universality, but even the simplest artificial NN models (e.g. the McCulloch-Pitts model) are universal.
One more minor thing (in agreement with what you said): there are some indications in the literature that there may be some neural network models that are greater than a Turing machine. This is subject to a lot of very shaky conditions. I'm not up to date with the latest on this front; but if someone claims the brain is more than a Turing machine, he'd better show some proof.
tl;dr answer: Here's a simple proof: If I describe a specific Turing machine to you, you can calculate its result, assuming you have enough paper or a big and complex enough brain.
Welcome to Slashdot. You're not supposed to read TFA.
to forge ahead all at once without trying to understand every step and the implications of every signal
That is, in fact, exactly what is being done.
Neural networks (both biological and artificial ones) "overparameterise", which basically means that they carry redundant information.
Due to this overparameterisation, it is mathematically impossible to assign an exact meaning to each connection weight*. Hence, it is impossible to assign exact meaning to a given signal coming into a neuron in the general case. You must consider multiple (often all) connections simultaneously.
All it would take is understanding of how to keep these cells alive and connecting
Incorrect. Neural network design is complicated. Not every network topology is suited to solving every kind of problem. Whether you use a threshold function or a continuous transfer function, and exactly which functions you use, impacts what problems you can solve and how well***. Granted, some designs are computationally universal, but this doesn't guarantee that the network will be easy to train, or that you won't need a trillion neurons to solve a given problem with a theoretically suitable design.
Furthermore, if the initial connection weights* are chosen poorly, the network may take a very long time to train or it may train towards a garbage solution.
You have heard that NNs are self-organising, but this is only true within certain limits. Neural networks are no panacea. Each network can approximate the solution to a given computational problem (or find the exact solution), assuming that there are enough neurons, assuming furthermore that the initial design was good, and also assuming that the training method is reasonable**. The brain is a collection of many different networks, of different topology and with many very different kinds of cells, that interact.
IANA neurobiologist, but IAA grad student studying artificial NNs.
* Biological neurons' inputs are a little more complicated than simply assigning a weight to each input and summing them, but the analogy holds. (Dendritic computation is still little understood.)
** For example, if you want a single hidden-layer feed forward NN to approximate a curve and training data is randomly distributed around the curve with a skewed distribution, the LMS algorithm is not ideal. (It may still be good enough, depending on specifics.)
*** Here I speak of artificial NNs. There are analogous differences in biological ones, such as the type of synapse (chemical/electric, axon-dentrite/axon-axon/dendrite-dendrite/axon-soma, unidirectional/bidirectional) and dentritic computation.