From: wfgodot@iquest.com (Michael Pierson)
Date: Mon, 13 Nov 1995 15:45:44 +0800
To: cypherpunks@toad.com
Subject: Re: coding and nnet's
In-Reply-To: <951110141713_18585763@mail04.mail.aol.com>
Message-ID: <nuRpw4B4GnSU088yn@iquest.com>
MIME-Version: 1.0
Content-Type: text/plain


-----BEGIN PGP SIGNED MESSAGE-----

In article <951110141713_18585763@mail04.mail.aol.com>, Atp3000 wrote:

>cypherpunks!
>I just resently got interested in coding and cryptography, and I was
>wondering if you could suggest 3 or 4 papers that adresses the aplication
>of neural networks in cryptography.



I just greped through "Abstracts in Cryptology" back to 1980 and
"Computer and Communications Security Reviews" up until September
1994 searching for the terms "neural" and "genetic."  The results below
show no evidence of research into the use of neural networks applied
to problems of cryptanalysis (in contrast to genetic algorithms). They
do however indicate their potential utility in such areas as transaction
and traffic analysis, as well as some types of biometric identity
authentication.

I wouldn't be surprised to learn of somebody like FinCEN using
neural-net systems to do pattern analysis on funds transfers and
the like, or the NRO or NSA investing research money into investigating
the usefulness of NNs for image processing or for scanning raw ELINT,
SIGINT or COMINT data.  In fact, I'd be quite surprised if they weren't.

While NNs may not be of direct relevance to cryptanalysis, I suspect
they are, or will be, of great relevance to the task of identifying
what communications should be cryptanalyzed or otherwise scrutinized.
Especially given the high volume of traffic our 'thinkpol' aspire to
be able to listen to.



 `The use of genetic algorithms in cryptanalysis'
 RA  Matthews, Cryptologia  v  17  no 2 (April 93) pp  187  - 201

 The author presents a genetic system for solving simple
 transpositions.  Candidate column orders are assessed for fitness using
 digram frequencies, and the best of them are used to breed a new
 generation of candidates, using column rotations and swaps as mutation
 operators.  The algorithm was useful at finding partial anagrams to aid
 manual solution.


 `Use  of a genetic algorithm in the cryptanalysis of simple substitution
 ciphers'
 R Spillman, M Janssen, B Nelson, M Kepner, Cryptolo gia v 17 no 1
 (Jan 93) pp 31 - 44

 The  authors report using a genetic algorithm to solve monoalphabetic
 substitutions. Randomly chosen keys were evaluated for fitness using
 letter and diagram frequencies, and the fittest candidates were then
 `mated' and subjected to mutation to provide the next generation of
 keys.  The `mating' operation used a selective crossover, in which the
 `best'character of each key was passed on.


 `Tearing up the rules'
 P Sampson, Banking Technology (Nov 93) pp 26 - 30

 Mellon Bank has installed a neural-net-based monitoring system which
 checks card transactions every two hours and looks for abnormal
 activity.  It is claimed to be substantially more effective than
 rule-based systems.


 `Cryptanalysis of knapsack ciphers using genetic algorithms'
 R Spillman, Cryptologia  v XVII no 4 (Oct 93) pp  367 - 377

 The author describes the design and use of a genetic algorithm to
 attack small trap do or knap-sacks and gives performance data to show
 that it functions 50 - 100 times faster than exhaustive search.


 `Neural net works: the way forward?'
 R Martin, Cards International no 99 (9/12/93)  p 9

 Neural networks are in use to detect abnormal card transactions in
 Mellon Bank and the Euro card Netherlands network, and both claimed that
 the system would pay for itself in 1-2 years.


 `Identity Verification using Weighted Personal Characteristics'
 Y Yamazaki, N Komatsu, M Tsuchiya, SCIS 94 paper 5C (in Japanese)

 The authors study whether characteristics of a person's handwriting may
 be used to authenticate that person.  They have used a neural net to
 distinguish persons successfully according to the angles between written
 strokes.


 `Comments on "Cryptanalysis of Knapsack Ciphers Using Genetic
 Algorithms" '
 F Rubin, Cryptologia v XVIII no 2 (April 1994) pp 153 - 154

 The author criticises Spillman's use of genetic algorithms to attack
 knapsack ciphers as ineffective against the kind of knapsacks
 actually proposed for non trivial cryptographic use.


 `Securenet: a network-oriented intelligent intrusion prevention and
 detection system'
 P Sprirakis, S Katsikas, D  Gritzalis, F Allegre, D Androutsopoulos,
 J Darzentas, C Gigante, D Karagiannis, H Putkonen, T Spyrou,  IFIP
 SEC  94  paper  E2

 This paper describes SECURENET, an intrusion detection system being
 built as an EC RACE project to protect integrated broadband
 communications. It uses a number of technologies, such as neural
 networks and secure distributed computation, to detect and classify
 attacks in real time.


 `Using  a  genetic algorithm for optimizing fixed polarity Reed-Muller
 expansions of Boolean functions'
 JF Miller, H Luchian, PV G Bradbeer, PJ Barclay , International Journal
 of Electronics v 76 no 4 (Apr 94) pp 601 - 610.

 The authors report a genetic algorithm which gets good sub-optimum
 Reed-Muller expansions of Boolean functions more quickly than previous
 techniques.



- -Michael


-----BEGIN PGP SIGNATURE-----
Version: 2.6.2

iQCVAwUBMKUpo9GJlWF+GPx9AQHuiQP8Dkp4o23j06TJCprZh7AYYEBH+IJQZzQW
wtlY+SfOzvcu+zEbQcj6q5G/feM7Ld1cmB/GK3u8qBPnM1OOQxxtvkR7dx//7svx
1P/E926zIOLZ3Q7FZqVLrsUdYJuKob+TXC2B4wMIQTFKXrJ0kNWEXvL4UpH26KkD
FEBXOQWnm8k=
=G5N2
-----END PGP SIGNATURE-----