Unbreakable Cryptography Theory...

Hey Guys-

I am a student in High School and over the past several months/years, I have been looking for a way to create a *nearly* unbreakable/unhackable cryptography system. Also, I have been experimenting with Neural Networks. So, here's what I have come up with for all you techno savy people...

To Encrypt Basic ASCII Text:
1) Begin with an aready organized Kohonen neural network...neurons are created for each character to be encrypted.
2) Your Public Key sets up the weight system of the Kohonen network.
3) Run the Kohonen network backwards to encrypt your data into a seemingly scrambled mess.
4) Use simple Perceptrons to look at the Hex code of the file and re-transform it into a whole different file...the Public Key would again be the weight setter.

You still with me? So to decrypt it, you would use a similar process. The nice thing about Neural Networks/AI with cryptography, is that they can sometimes be smarter than standard algorithms.

If you guys would, please comment on how I could improve upon this theory. Also, if you don't have a good grasp on Neural Networks, visit http://www.generation5.org/essays.shtml#nn. His essays should allow you to become fully familiar with the general information about Nerual Networks.

*Thanks!:)*

-BorgCubes

Umm,

Senor Borg thou is missing an "l" from your link.

So what you're saying is that you can use a neural network to do asymmetric encryption... cool but:

1. How fast is it and does it use much memory?
2. Has it been extensively cryptanalysed?
3. Is any of the maths behind it even vaguely provable?

I doubt whether (1) is very important, because asymmetric encryption is very rarely used for a large volume of data (symmetric keys normally used instead, they are much faster), although if it needs like 10Mb of memory for each operation that would be a problem (think how slow an SSL handshake would be, like 10Mb of memory on the server for each session, even if it was freed after handshake and symmetric encryption used thereafter)

Right now, I am not too worried about how long it takes it to encrypt/decrypt because a simple Kohonen program I have takes about 1 minute to sort out simple data. The memory will only become an issue when you have a TON of bytes in the file, cause one neuron would be created for every byte that the file has. On most standard machines with at least 512mb of memory would be fine...if you wanted to do something really big, the program could do it in small pieces (which would take longer) or you might have to find a better machine to do it on...

...hopefully, over the next few days, I'll have a peliminary program written to disorganize/encrypt data!

:)

-BorgCubes

I would not worry about the performance on a large file, because it is highly unlikely anyone would want to use asymmetric encryption for large data.

Normal protocols and applications use an asymmetric cipher only to encrypt a key for a symmetric cipher (which is very unlikely to be >256 bits), which is then used to encrypt the data.

Remember that bruting the encryption is basically always possible if you know the encryption method (without knowing how to decrypt) and the encrypted data. However if the encryption method is slow the bruteforcing might take almost a year... but it IS possible.

Quote:

---

Originally posted here by Haikka
Remember that bruting the encryption is basically always possible if you know the encryption method (without knowing how to decrypt) and the encrypted data. However if the encryption method is slow the bruteforcing might take almost a year... but it IS possible.

---

Well unless you're a government or corporation with crazy powerful computers, no home user or computer guru will be able to brute force many of today's encryption schemes within their lifetimes. Unless you get incredibly lucky and start with a key almost identical to the original and get it within the first few thousand to million tries. Most encryption is broken by a flaw in the algorithm making it relatively simple if you understand what you're doing, relatively could mean you need a PhD in advanced mathematics, or by discovering the key through alternate methods. When you have a key that's 128-bits long that means there's 2^128 combinations... and that's becoming the standard key size in most encryption these days, or at least the smallest used in most cases. If you could make your home computer run through all those possibilities within a year while still knowing that you broke it when you do, unless you tell your program how to tell if you broke the encryption there's no real way for it to know to stop.

I mean a bruteforcing like this:



Encrypted string: sasfgfiealskdmlasieeqskmdapwowmqwfdmafoewpq
The program starts encrypting loads of strings:
a
b
c
...
hello, this is a tesr -> asddamopemlaspfomsföasmfpaeomsgalfmadfsoöae
hello, this is a tess -> alsdaispodmaoifmsdlkmfapowepqoweialksömdpofa
hello, this is a test -> sasfgfiealskdmlasieeqskmdapwowmqwfdmafoewpq [same as the known encrypted string]

-> bruteforcing complete <-

Oh well it would be different if the encryption wouldn't use a fixed key...

Yeah and not to mention that doing it that way would take even longer because what if it's a 50 page report that's encrypted? By the time you figured out the original text it'd be out to the public anyway or it would be completely obsolete. And not to mention if the key can be anything like most schemes you'd have to not only figure out the original text but then figure out what key was used. And then you could end up with multiple possible solutions... one key will work with one possible text yet another key will work with a completely different text.. and if it's data then it would make it even harder to distinguish.

Thanks for all the info and stuff...now onto another question:...

...would anyone like to help me program the neural networks to accopmplish this task? Once you have your basic structure down, it will be much easier to implement standard cryptography algorithms/genetic algorithms into the code.

So if anyone is up for a challenge, and if anyone know anything about NN's, I would greatly appreciate your help!

*Thanks* :)

-BorgCubes

So in the end, your "unbreakable Cryptography" isn't unbreakable. It's simple unfeasible to break due to time\ machine power constraints.

Quote:

---

Originally posted here by Haikka
I mean a bruteforcing like this:



Encrypted string: sasfgfiealskdmlasieeqskmdapwowmqwfdmafoewpq
The program starts encrypting loads of strings:
a
b
c
...
hello, this is a tesr -> asddamopemlaspfomsföasmfpaeomsgalfmadfsoöae
hello, this is a tess -> alsdaispodmaoifmsdlkmfapowepqoweialksömdpofa
hello, this is a test -> sasfgfiealskdmlasieeqskmdapwowmqwfdmafoewpq [same as the known encrypted string]

-> bruteforcing complete <-

Oh well it would be different if the encryption wouldn't use a fixed key...

---

This is mostly inaccurate. Most modern algorithms 'scramble' they key and the plaintext between or during each round of encryption. The only real way the example you have provided would work is if the system used a mono-alphabetic system. With any type of poly-alpha system you will not know if you get close to the correct answer, simply because the set of answers to the system can be (with some 0x00 padding) "hello, this is a test", "hi, you evil evil person", "what is this about?", and any other message of the correct length.

As far as this cryptosystem goes, for some reason people always walk up to neural networks for crypto, most of the time trying to make a ANN that will break crypto (which isn't as easy as it looks because there is only one right answer and the key changes everytime, thus resetting all the wieghts to 0, so it's almost pointless to try unless you find a 'shortcut'). This is definately a 'creative' thought...I'd like to see an implementation of it as well as a white paper on the algorithm design so cryptoanalysis would be possible.