Slashdot Mirror
Visual Exploration of Complex Networks
jweebo writes "Seed magazine has a story on complexity, and how it can be visually represented with fascinating results. From the article:
'Complexity is everywhere. It's a structural and organizational principle that reaches almost every field imaginable, from genetics and social networks to food webs and stock markets ...Collected here are a few of the many intriguing, and often beautiful, images that illustrate how the whole is more than the sum of its parts.'"
The article is a little short, I would have liked more more more!! :-)
May I suggest Information Architecture from Peter Morville. He is also co-founder and president of the Information Architecture Institute.
May I also suggest taking a look at Prefuse, an open source project to interactively vizualize organized information (still in beta however).
Animoog.org
People may be more complex than a neuron, but not nearly as complex as the total and their possible interactions. With billions of neurons, each interacting with 1,000 to 10,000 others the possible configurations are enormous, yet most we do not act in such a wildly differing manners. I don't know much about chaos theory, but it mentions something along the lines of the sinple behaviour of complex systems and the complex behaviour of simple systems. Thinking deeper, I am not sure which category this falls into.
I don't make predictions, and I never will.
Something along similar lines is Frank van Ham's work on visualizing large state spaces. He's generated some neat visualizations of complex transition systems associated with various protocols.
I judt got a nre Kinesis keybiartf so please excusr ant egregiou typos.
-
pixels on the display: 2 million or so.
- insufficiency of the clustering algorithms: showing one pixel per node and random placement, or placement by DFS traversal? for trees, or for graphs where classification is the primary concern, then tree-map or "Csoft" views scale relatively well in this regard, but what about for more general problems?
- implementations (or algorithms) that don't scale: e.g. graphviz uses n^2 (n=#nodes) space for its graph layout!
one must always think about the summarization criteria: what aren't you going to show? how will you indicate that summarization has occured? how do you denote drill-down capability? what will the form of drill-down be? what heuristics should you use to selectively deaggregate, in order to highlight potentially interesting subgraphs? for large-scale info, this is as important as what you will be showing, and how it will be shown! for our stuff, we have graphs with tens of millions of nodes.Yes, he got his PhD at a very young age, BUT... He seems to think that he doesn't need to adhere (anymore) to the principle of how scientific work is normally communicated to others, i.e., through peer-reviewed journal / conference publications. Instead, he puts all of his work in a book that isn't peer-reviewed by fellow scientists before being published, giving him the opportunity to make all kinds of unsubstantiated claims.
The fact that he doesn't cite others whenever this would have been appropriate makes the uninformed reader think that Wolfram actually came up with a lot of the material in the book himself, i.e., that it actually represents "A New Kind of Science", but fact of the matter is that a lot of what he presents has already been researched in one form or another by other scientists.
He tries to refute the criticism regarding his "non-peer-reviewed, non-citing"-approach by simply claiming that his "New Kind of Science" doesn't fit within the normal way of doing research; which sounds like an easy way out to a lot of other scientists.
...than a thousand words.
Really, you'd be amazed at how even the simplest graphical interpretation of complex data can really show up points of interest. And it's not difficult to see why: Humans' primary sense is visual and we have evolved some seriously complex neural algorithms to interpret visual data.
A simple graph is a case in point. Now take a large amount of complex data and apply just about any process you care to name to present a graphical representation and you can easily see the overall picture.
A very simple example which illustrates statistical clustering. Even with totally random numbers, you *will* find islands of apparently significant populations. This is a common counter-claim to action groups who claim, say, a correlation between mobile 'phone masts and incidents of child leukaemia*. Anyway:
Generate a stream of random numbers and assign a symbol for n = 0.5, display the symbols in a grid and, hey presto! Look at those clusters!
On a more positive note:
We often use graphical representation in our work. This ranges from CTK representations of molecules we're looking at (xlation - pretty pictures with balls and lines) to grid based colour indexed representation of multi-dimentional data sets. In each case the point is to present data in a way that we humans can quickly spot potential areas of interest and get a "feel" for the data we're looking at.
It's all good stuff. (Sometimes very pretty, too)
* Actually, this is a good example of why I'm always wary of purely statistical "proofs". In this case the *science* (ie. proposed mechanisms for this) don't hold up to current understanding.