Christian L. Staudt

• Dr. rer. nat. (=PhD)
• Analyst at https://clstaudt.me/

 A line of research (e.g. [1]) has considered the question of self-similarity and self-dissimilarity in complex networks.

Very generally, a self-similar object is similar to a part of itself. A natural application of this definition to networks would be to say that a network is self-similar if its graph is composed of subgraphs that are structurally similar to the graph as a whole. Still, this is not yet a precise, quantifiable definition - for instance, what does it mean to say that the graph is "composed" of subgraphs?

Please comment and discuss.

[1]: http://arxiv.org/abs/cond-mat/0503078

Humans are good at spotting patterns visually, which can be used for exploratory data analysis, but we can be overwhelmed by visual complexity. The usual drawing methods for networks result in pretty pictures that illustrate "complexity", though arguably not much else [1]. Typical graph drawing methods are inadequate for large networks with thousands of edges (and many interesting data sets are considerably larger than that), as they mostly result in "hairballs".

It seems to me that in order to draw large networks meaningfully, we have to go beyond the typical node-link diagram with some type of force-directed layout. Ideas in this direction are hive plots [2] and hierarchical edge bundling [3].

Which visualization methods do you know that are able to expose important features of large complex networks more effectively?

[1]: http://www.hiveplot.net/img/network-communities.png

[2]: http://www.hiveplot.net/

[3]: https://mbostock.github.io/d3/talk/20111116/bundle.html

There is basically no limit to the phenomena that can be modeled and analyzed in terms of complex networks - entities and their relationships between which can be represented as the nodes and edges of a graph, and which form a non-trivial pattern. So let's make this a small survey:

Nowadays complex network data sets are reaching enormous sizes, so that their analysis challenges algorithms, software and hardware. One possible approach to this is to reduce the amount of data while preserving important information. I'm specifically interested in methods that, given a complex network as input, filter out a significant fraction of the edges while preserving structural properties such as degree distribution, components, clustering coefficients, community structure and centrality. Another term used in this context is "backbone", a subset of important edges that represent the network structure.

There are methods to sparsify/filter/sample edges and preserve a specific property. But are there any methods that aim to preserve a large set of diverse properties?

In the wake of the Snowden leaks, the public has learned that the sheer technical scale of surveillance goes beyond what has usually been thought possible, such as a "full take" of internet communication. The picture emerges that intelligence agencies are prominent users and drivers of "big data" technology. But from the media reporting, capabilities remain rather unclear, certainly also because of the variety of different surveillance programs, the complexity of the technology, and of course secrecy.

I would like to know more about what the technologies employed are and how far technical capabilities go. Since it may be a vast topic, let me narrow the question down: I would like to learn about data analysis capabilities, not about the use of cryptology or computer security research for surveillance. The focus is on mass surveillance rather than spying on specific target persons. I am interested in efforts to make society as a whole "machine-readable" by using techniques such as data mining, machine learning, social network analysis etc.

Especially interesting are things like:

- the intent to build a comprehensive social graph of large population segments [1]

- the use of social media and other data streams to anticipate political events such as protests, civil unrest etc., e.g. [2]

Can you point to publications? Work done in the service of surveillance and contributions of computer scientists to the critical debate around the issue would be equally interesting.

[1]: http://qz.com/129425/how-the-nsa-builds-social-graphs-of-americans-with-phone-email-and-location-records/

[2]: http://arxiv.org/abs/1402.7035