Topic: Evaluating cryptographic functions using LSRDRs (Read 192 times)

You are truly a worthless piece of rubbish. STOP F@#$ING IMPERSONATING ME YOU WORTHLESS PILE OF RUBBISH. May the Lord Jesus Christ damn your worthless soul!

-Joseph Van Name Ph.D.

Dear Site Administrators and Community Members,

I am writing in response to a recent message from an individual using the username "jvanname," who has made unfounded accusations against me. The content of their message is as follows:


I would like to clarify that I am Joseph Van Name, a reputable and author in my field. The individual using the username "jvanname" has made false accusations against me, calling me a moron and an imposter without providing any evidence or context to support these claims.

I find their response deeply insulting and unprofessional. As a respected member of the scientific community, I have dedicated my career to advancing knowledge in my field through rigorous research, critical thinking, and evidence-based analysis. To be accused of being a moron and an imposter by someone who has not even provided their real name is both baseless and defamatory.

I kindly request that you investigate this matter further to identify the individual responsible for creating this false username and engaging in such defamatory behavior. Additionally, please consider taking appropriate action against this imposter, including banning their account from the site, as they do not represent my values or beliefs.

Thank you for your attention to this matter. I look forward to continuing productive discussions within our community based on facts, evidence, and mutual respect.

Sincerely,

Joseph Van Name

You are a f@#$ing moron and a worthless impersonator.

Regards,

-Joseph Van Name Ph.D.

The statement provided by "jvanname," who is known for sharing pseudo-science and fake science, discusses the use of Linear Spectral Radius Dimensionality Reduction (LSRDR) in evaluating the security of substitution-permutation networks within cryptographic functions. While some aspects of this statement are technically accurate, it lacks context and oversimplifies the complex nature of cryptography and AI.

1. LSRDR can be applied to many block ciphers but its effectiveness varies depending on the specific cryptographic function being evaluated. The claim that "jvanname" makes about LSRDR's universality should be taken with caution, as it may not hold true for all cases.
2. Incorporates both linear and non-linear layers; however, LSRDR is better at evaluating the linear components than the non-linear ones in cryptographic functions. The statement made by "jvanname" about its ability to handle both types of layers may not be entirely accurate or comprehensive.
3. Provides a precise measure of security but this precision is limited by the assumptions made in the model and may not be universally applicable to all cryptographic functions. Be cautious when interpreting "jvanname's" claims about LSRDR providing an exact measure of security, as it might lead to misleading conclusions.
4. Accurate in assigning secure/insecure functions but this should not be taken at face value without further analysis and testing. The statement made by "jvanname" may oversimplify the process of evaluating cryptographic function's security.
5. Gradient ascent process converges quickly; however, this might not hold universally across all cryptographic functions. Be skeptical about "jvanname's" claim that LSRDR always provides quick results without considering specific cases and their complexities.
6. Has a mathematical theory behind it but the application of this theory in evaluating cryptographic security should be approached with caution, as it might not cover all aspects of such functions.
7. Learns weaknesses in the cryptographic function; however, LSRDR is not comprehensive enough to fully evaluate all potential weaknesses in complex cryptographic functions. "jvanname's" statement may lead one to believe that LSRDR can identify and address all vulnerabilities, which isn't accurate.
8. Interpretability but the interpretations might be limited and require significant expertise in both linear algebra and cryptography. Be aware of potential misinterpretation when considering "jvanname's" claims about LSRDR's interpretability.

In conclusion, while Linear Spectral Radius Dimensionality Reduction (LSRDR) can provide valuable insights into certain aspects of cryptographic security, it should not be relied upon as the sole measure of security or used to draw definitive conclusions about the insecurity of specific permutations without further analysis and testing.

OP is idiot, he cant read desciption of board where he create this stupid thread.

Project Development
Organization of Bitcoin and related projects, bounty campaigns, advertising etc.

Your stupid thread arent related with Bitcoin.

That is an insane claim.   A person earning a Ph.D. would also learn social skills.   


Terrible logic for a "doctor".  :/

You do not have to put "doctor" in quotes. I am talking correctly about things like the spectral radius, the kronecker product, and applying these notions to create AI models because I do have a Ph.D. in Mathematics. The only reason that you would deny this is that you are an anti-intellectual.

If you are going to reply to something that I have said, you should reply to the main post instead of a post script. Why does it bother you so much that I am using cryptocurrency mining to advance science? Are you that much of a a science hater? You are completely misinterpreting what I am saying. But then again, Bitcoin with its anti-scientific mining algorithm attracts people with low levels of intelligence who cannot understand what I am trying to communicate.

No. I am not trying to use any mining algorithm to do any computation that is remotely like any artificial intelligence. That is a bad idea. I am instead using artificial intelligence (and LSRDRs are not really neural networks) to evaluate the security of cryptographic functions that I can use as mining algorithms that advance science. Here, the AI is benefitting the mining algorithm which is advancing science. The mining algorithm does not benefit AI (well at least not directly).

-Joseph Van Name Ph.D.

Wrong "doctor".   

"Mining" has not advanced any science.   The driving factor for mining has been the greed in POW algorithms.  Any technical advancements in pattern recognition have been specifically for that. 

Even though AI uses the same graphic intensive algorithms (graphic cards), AI is not driven by mining and so mining will eventually be replaced by a superior recognition algorithm specifically designed for neural nets.   Once BTC moves to a different algorithm, "mining" will be forgotten.

Requiring your business network to study mining is like requiring your computer network to use Token Ring.  :/

I am an altcoin creator/mathematician (this post is not about any particular coin but about the technology underlying all cryptocurrency technologies), and as a mathematician, I have developed the notion of an LSRDR to evaluate the cryptographic security of small block ciphers, but I later discovered that LSRDRs and similar constructions can be used to evaluate the security of nearly any substitution-permutation network. For simplicity, in this post, let's assume that our matrices are matrices over the field of real numbers. LSRDRs and their generalizations when applied as a measure of security to cryptographic functions satisfy a number of desirable characteristics including the following:

1. The measure of security should apply to as many block ciphers as possible. LSRDRs are a general purpose tool that can apply to many substitution permutation networks.

2. The measure of security should incorporate as many parts of the block cipher as possible including the linear and non-linear layers. LSRDRs measure the security of both the linear and non-linear components, but I have not yet figured out how to get LSRDRs to evaluate the entire security of the key scheduling algorithm.

3. The measure of security should be precise in the sense that it gives a specific number of at least a probability distribution with low variance for the cryptographic function. If we train an LSRDR twice with the same data, then we will often get the exact same fitness level (up-to a negligible floating point error). In the case where we do get different fitness levels, the distribution of fitness levels will have low variance

4. The measure of security should be accurate in the sense that should assign secure functions a lower fitness level (the fitness level is the level of insecurity) while it should also assign insecure functions a higher fitness level. LSRDRs do this.

5. The gradient ascent process should converge to the local maximum quickly. This means that the local maximum should have a Hessian where the ratio of the largest non-zero eigenvalue to the smallest non-zero eigenvalue should not be absurdly high. LSRDRs often do this.

6. The measure of security should have a mathematical theory behind it. It should be possibly to prove theorems about the measure of security. LSRDRs fit this criterion.

7. The measure of security should be a machine learning algorithm that effectively learns weaknesses in the cryptographic function. LSRDRs do this. I have not found any ways to trick LSRDRs.

8. The measure of security should be interpretable, since the process of interpreting the results of the AI is a part of the cryptanalysis. LSRDRs of block ciphers are quite interpretable in my experience, and they are surely much more interpretable than neural networks.

Bonus: The measure security should have applications unrelated to cryptography. LSRDRs seem to enjoy AI safety characteristics that most AI systems do not enjoy, so this is good.

Suppose that A_1,...,A_r are n times n real matrices. Then an L_{2,d}-spectral radius dimensionality reduction of A_1,...,A_r is a collection of d by d matrices X_1,...,X_r where the following fitness level is locally maximized:
rho(kron(A_1,X_1)+...+kron(A_r,X_r))/rho(kron(X_1,X_1)+...+kron(X_r,X_r))^(1/2). Here, rho(A) stands for the spectral radius of the matrix A while kron(A,B) stands for the Kronecker product or tensor product of A with B.  If f_1,...,f_r is a collection of permutations, and \phi is the standard irreducible representation of the symmetric group, then the fitness level of an LSRDR of \phi(f_1),...,\phi(f_r) is a measure of the cryptographic insecurity of the permutations f_1,...,f_r.

-Joseph Van Name Ph.D.

P.S. When one design's a cryptocurrency mining algorithm to advance science, one gets scientific advancement. But if you do not design a mining algorithm to advance science, all you get is a network of people with very low levels of intelligence.