Darknet Mining for Proactive Cybersecurity Threat Intelligence
They are using machine learning to comb the darknets, capturing about 300 threats a week.
About 90% hack application and backdoor recognition, that is for sale, and about 80% hacker forum vulnerability identification.
"These threat warnings include information on newly developed malware and exploits that have not yet been deployed in a cyber-attack"
[Link] An exact mapping between the Variational Renormalization Group and Deep Learning]
An exact mapping between the Variational Renormalization Group and Deep Learning by Pankaj Mehta, David J. Schwab
Deep learning is a broad set of techniques that uses multiple layers of representation to automatically learn relevant features directly from structured data. Recently, such techniques have yielded record-breaking results on a diverse set of difficult machine learning tasks in computer vision, speech recognition, and natural language processing. Despite the enormous success of deep learning, relatively little is understood theoretically about why these techniques are so successful at feature learning and compression. Here, we show that deep learning is intimately related to one of the most important and successful techniques in theoretical physics, the renormalization group (RG). RG is an iterative coarse-graining scheme that allows for the extraction of relevant features (i.e. operators) as a physical system is examined at different length scales. We construct an exact mapping from the variational renormalization group, first introduced by Kadanoff, and deep learning architectures based on Restricted Boltzmann Machines (RBMs). We illustrate these ideas using the nearest-neighbor Ising Model in one and two-dimensions. Our results suggests that deep learning algorithms may be employing a generalized RG-like scheme to learn relevant features from data.
To me this paper suggests that deep learning is an approach that could be made or is already conceptually general enough to learn everything there is to learn (assuming sufficient time and resources). Thus it could already be used as the base algorithm of a self-optimizing AGI.
Request for feedback on a paper about (machine) ethics
I have written a paper on ethics with special concentration on machine ethics and formality with the following abstract:
Most ethical systems are formulated in a very intuitive, imprecise manner. Therefore, they cannot be studied mathematically. In particular, they are not applicable to make machines behave ethically. In this paper we make use of this perspective of machine ethics to identify preference utilitarianism as the most promising approach to formal ethics. We then go on to propose a simple, mathematically precise formalization of preference utilitarianism in very general cellular automata. Even though our formalization is incomputable, we argue that it can function as a basis for discussing practical ethical questions using knowledge gained from different scientific areas.
Here are some further elements of the paper (things the paper uses or the paper is about):
- (machine) ethics
- (in)computability
- artificial life in cellular automata
- Bayesian statistics
- Solomonoff's a priori probability
As I propose a formal ethical system, things get mathy at some point but the first and by far most important formula is relatively simple - the rest can be skipped then, so no problem for the average LWer.
I already discussed the paper with a few fellow students, as well as Brian Tomasik and a (computer science) professor of mine. Both recommended me to try to publish the paper. Also, I received some very helpful feedback. But because this would be my first attempt to publish something, I could still use more help, both with the content itself and scientific writing in English (which, as you may have guessed, is not my first language), before I submit the paper and Brian recommended using the LW's discussion board. I would also be thankful for recommendations on which journal is appropriate for the paper.
I would like to send those interested a draft via PM. This way I can also make sure that I don't spend all potential reviewers on the current version.
DISCLAIMER: I am not a moral realist. Also and as mentioned in the abstract, the proposed ethical system is incomputable and can therefore be argued to have infinite Kolmogorov complexity. So, it does not really pose a conflict with LW-consensus (including Complexity of value).
Robot ethics [link]
The Economist has a new article on ethical dilemmas faced by machine designers.
Evidently:
1. In the event of an immoral decision by a machine, neural networks make it too hard to know who is at fault--the programmer, the operator, the manufacturer, or the designer. Thus, neural networks might be a bad idea.
2. Robots' ethical systems ought to resonate with "most people."
3. Proper robot consciences are more likely to arise given greater collaboration among engineers, ethicists, policymakers, and lawyers. Key quotation:
Both ethicists and engineers stand to benefit from working together: ethicists may gain a greater understanding of their field by trying to teach ethics to machines, and engineers need to reassure society that they are not taking any ethical short-cuts.
The second clause of the above sentence is quite similar to something Yudkowsky wrote, perhaps more than once, about the value of approaching ethics from an AI standpoint. I do not recall where he wrote it, nor did my search turn up the appropriate post.
Free Online Stanford Courses: AI and Machine Learning
Stanford has decided to offer a few classes online, for free. These include Artificial Intelligence and Machine Learning. The classes include videos of the same lectures that the Stanford students received, quizzes, homework, and exams that are graded automatically. They start on October 10.
I'm guessing that more than a few LWers will sign up for these. How many people would like to form a study group? Should we just have a discussion thread for it, or is there a better option?
Shane Legg's Thesis: Machine Superintelligence, Opinions?
I searched the posts but didn't find a great deal of relevant information. Has anyone taken a serious crack at it, preferably someone who would like to share their thoughts? Is the material worthwhile? Are there any dubious portions or any sections one might want to avoid reading (either due to bad ideas or for time saving reasons)? I'm considering investing a chunk of time into investigating Legg's work so any feedback would be much appreciated, and it seems likely that there might be others who would like some perspective on it as well.
= 783df68a0f980790206b9ea87794c5b6)
Subscribe to RSS Feed
= f037147d6e6c911a85753b9abdedda8d)
Since risk from individual SNP's 'should' not be aggregated to indicate an individual's risk based on multiple sources of evidence, how are the magnitudes for genosets determined?. Can bayes or another method be used to interpret a promethease report?
Even genetic epidemiology textbooks seem pessimistic: about the usefulness of the genetic research underpinning precision medicine:
The references in question are about the impact of population stratification on genetic association studies. That doesn’t seem to substantiate such a broad stroke about the non-replicability of genetic epidemiology. I don't know what to make of these findings.
Here is a link to a screenshot of those references
It suprises me that entrepreneurial machine learning analysts don’t beg for genetic research to identify how combinatorial patterns of genes to be able to characterise individual risk. It seems like if/once they can get hold of that information, the sequence from genetic science to consumer actionable health information is bridged. So where are the 'lean gene learning machine' startups? I certainly don’t have the lean gene to do it myself. I don’t know machine learning.
Regulatory issues seems like the biggest hurdle. To the best of my google-fu, 23andme doesn't even disclose what it's 'Established Research' genes are. So, once regulatory hurdles are surmounted, lots of useful research will flood out.