Suppose that the NSA had announced the possession of an efficient factorization algorithm. The cryptology community, after recovering from the initial shock, would demand to see the algorithm and verify it. This request, however, could not be satisfied since the algorithm would probably be classified as top-secret information.
The question now is: What can we do? The answer is: We need Zero-Knowledge Proofs of Computational Power. 
I just finished reading the paper “Secure Provenance: the Essential of Bread and Butter of Data Forensics in Cloud Computing” presented at the ASIACCS 2010. In the paper, a new secure provenance scheme based on bilinear pairing techniques is proposed. I really enjoyed reading the paper, however, I didn’t realized at first that the authors focused rather on the cryptographic system (bilinear pairing techniques) than on the practical applicability of such techniques in the field of cloud forensics. For instance, in section 2.1, the Service Provider is expected to be “… trustable and has significant resources and monitors live cloud computing systems”.
And this preliminary is the main issue, if you ask me. Cloud Service Providers (CSPs) cannot be seen as trustable. Especially this aspect makes the whole “cloud” so complex and difficult. It is slightly narrow-minded to consider only the user as “bad”. The CSP, intentionally or not, also states a huge problem concerning the security of customers’ data. We will see if some solutions will be found for this problem …
… is the title of this interesting and informing blog posting about the search for the holy “crypto” grail. 
It also explains the homomorphic encryption possible with textbook RSA that can easily be understood also by non-cryptographic affiliated people.
You guys know this already? It’s one of the more useful pages of the Springer Verlag – and yes, you don’t have to pay for that service. I already searched for one or two equations I currently have to work with and I pretty much succeeded. Try it yourself!
I found some nice statistics about the used email addresses and TLDs of the carders.cc customers. As you might have noticed, the dump is out there in the wild, available for download.
These few words by a math professor from the UCSD are quite worth to be mentioned.
If you come across a problem in your area of research, there are quite often different approaches and hence different ways to solve the problem. I currently have to differentiate between the following two ways:
Approximate the solution to the right problem
vs
Use the optimal solution to the approximate problem
I like this differentiation and I will probably decide to use the first approach. 
I’m currently quite busy and hence have almost no time for doing “interesting” things, however I recently stumbled upon this webpage which contains a lot of interesting videos: http://videolectures.net/
Especially in the field of computer science, a lot of high level speakers were recorded. I recently watched a lot of videos about Monte Carlo Algorithms and therefore can highly recommend the talk by Arnaut Doucet. But you should also check out the other speakers also …
A swedish researcher named Thomas Schön has a very profound archive of papers related to nonlinear estimation approaches. A lot of pdfs are free for download on his webpage.
I currently read the following two papers:
http://www.control.isy.liu.se/student/graduate/dynamic_vision/Lectures/NLEstimation.pdf
and his phd thesis:
http://www.control.isy.liu.se/research/reports/Ph.D.Thesis/PhD998.pdf
A company named Vose Software did a great job in summarizing lots of different probability distributions and putting them into one huge 170 pages pdf. If you work on probability stuff, this can be quite useful.
Seriously, I had the same idea a few years ago, but obviously I was too lazy to make it run. Adrian Dimcev is currently working hard on this project and I really look forward to diving deep into the results. Get more information on Adrian’s blog and have a look at the beta results here.
The slides for the workshop “Statistical and Learning-Theoretic Challenges in Data Privacy” are online now so if you wonna have a look, go ahead! Jonathan Katz is currently also blogging about the workshop. As the title says, the workshop focuses on theoretical challenges, so don’t expect too much practical cover.
On Colin Percival’s Blog I recently found an nice article about howto hack the Amazon S3 SLA. The article explains which kind of actions should be taken in order to safe money. Although it is a little bit “theoretically” based, it is really worth having a look at.
To install R for Mac OS X is quite easy: Chose one of the mirrors of the R-Project, download the Mac OS X version and install it.
After the successful installation, you want probably edit the first file. For this purpose, you chose the editor Textmate which can handle a lot of languages, unfortunately not R by default. This means, you have to install an additional bundle for that reason. On this blog, a small and fast explanation shows how to do this. It’s really only a question of 1-2 minutes.
Afterwards, you can enjoy the real power of Textmate with this small R sample.
DisSev <- c(0,12,15,34,23,28)
RelYield <- c(12,23,34,24,16,17)
plot(x=DisSev, y=RelYield, xlab=’Disease Severity (%)’,
ylab=’Relative Yield (%)’)
If you’re curious about the result, here it is after pressing CMD-R:
Awesome! Nice visualizations within a couple of minutes work.
PS: Sweave is the name of the dynamic interface between LaTeX and R.
websequencediagrams.com offers an interesting feature allowing you to draw a sequence diagram in roughly under a minute.
note over A,B: text1
note left of A: text2
note right of A
multiline
text
end note
leads to the following illustration:
Cool stuff – fast and easy!
This paper is a very fine example for what particle filters can be used. Take for instance a first-person shooter like “Counterstrike” with bots as adversaries. The authors of the paper focused especially on games where hunters, or searchers, spread throughout a map passively guarding regions or actively searching for a player. The player wishes to accomplish his set of goals without having his exact position discovered by the searchers.
As the searchers do not communicate continuously, the player can disable a searcher that has discovered significant evidence of his position, preventing the searcher from communicating. The other searchers retain their own inferences and can continue to collaborate, while the state held by the disabled searcher is unavailable. Should the disabled searcher recover, he can rejoin the searchers and pass his inferences to the other searchers. The player could also feed false information to the searchers (by causing a noise, for instance), who would act on and communicate this misinformation. These induced errors in the model are eventually corrected by the searchers investigating and observing the true state, but would provide a temporary distraction.
Quoted from Particle-based communication among game agents
Currently I dig deep in the fields of stochastic theory. More precisely, I read a lot about Sequential Monte Carlo methods (SMCs) and Particle Filters in general. For non-mathematicians like me, it’s really tough stuff. But I hopefully can handle it with a lot of efforts.
I just finished reading the work by Villaverde and Ramirez called “Estimating dynamic equilibrium economies: linear versus nonlinear likelihood” which originates more from the economical field of research. However, I consider this work as a good step for getting the very basics of SMCs in general. I really learned a lot from this work. The authors distinguish linear (Kalman) and non-linear (SMCs) filters and discuss the advantages for estimating dynamic equilibrium models. Sounds crazy and it quite is – especially the evaluation algorithms are difficult to understand.
For the better understanding of the methodology, I visualized the state model which is similar to the one used in the paper:
This figure shows the context of the different states. The transition function is referred to as g() and the measurement function is named h(). Obviously, the measurement of the state vector x is influenced by the noise vector v. Therefore, the observation z is only an approximation of the state vector x. Any comments so far?
The intention to use SMCs is to get information about the real state vector – not the noise-corrupted measurement vector. So SMC methods have the intention to estimate the current unknown probability density of the state space for derivating the most probable state of a dynamic system. This means, SMC methods can be used to estimate the state in a dynamic process in which only the crucial disturbance variables are known. In most of the cases, observations can only be performed on a partly basis which means that there are a lot of hidden states within the system.
It is now possible to make assumptions about the real forensic state vector with the help of SMCs. PF methods make use of large sets of random samples, so called particles, for approximating the sequence of probability distributions.
Finally, I have to admit that it’s kind of black magic to handle such stuff. You cannot understand this topic with common sense – or at least this is not possible for me. I still try to find the “golden thread” in this field of research. There are so many parameters which have to be taken into account – so feel free to correct me – I would really appreciate it!
