On Fri, Sep 5, 2014 at 4:04 PM, Alberto G. Corona <[email protected]> wrote:
> Telmo: > > Really impressed by your work. > Thanks Alberto! > > The generative rules however must have a low descriptive level in terms of > lengths of graphs number of connections etc. Am I right? > Well, the generators simply give you the likelihood of a connection between two given nodes. Iteratively, you can use the generators to produce a graph with n nodes and m connections. So you can use the same generator to produce networks of different sizes, but you do have to predefine the size. > > To earn the status of artificial scientist, How these low level terms can > be "elevated" to tell something meaningful about the concrete problem > studied? > Well, it's "shut up and calculate" type of scientist. Could an AI go all the way and attempt an interpretation? I think so, but unfortunately I don't have the algorithm yet... In the blog post I make an effort to interpret the equations. To try to answer your question, consider the "political blogs" case. The expression is: w(i, j) = exp(4 - 2d) One possibility that this raises: maybe we can explain bi-partidarism as the simple outcome of social contagion. You ran this generator (for the network size of the real case) and you get two communities with a small interface -- just like the political blog network discussing the elections in 2004. It proves nothing, of course. But it hints at something. > > For example what the generator rule found for Facebook tell about Facebook? > This one seems to match intuition very well. I tells us that people prefer to connect to popular people, and that, at the same time, social cliques from the outside are transferred into facebook. > > I mean, to find a low level generative rule is impressive but are there > more? > There are more. The method simply looks for the simplest explanation. It applies Occam's razor, and we found evidence that, in doing that, it tends to converge on similar explanations. It is of course possible that the more complex explanation is the correct one, but here we are faced with the exact same problem that human scientists face. The best we can do is assume that the simpler explanation is more likely. Cheers, Telmo. > > > > > > 2014-09-05 14:20 GMT+02:00 Telmo Menezes <[email protected]>: > >> Hi all, >> >> Since people have been talking about AI, creativity etc., I take the >> liberty of doing a bit of self-promotion. >> >> My paper "Symbolic regression of generative network models" has finally >> been published and it's open access. Here's a blog post about it: >> >> >> http://www.telmomenezes.com/2014/09/using-evolutionary-computation-to-explain-network-growth/ >> >> and the direct link: >> >> http://www.nature.com/srep/2014/140905/srep06284/full/srep06284.html >> >> The idea of this work is to use genetic programming to evolve plausible >> bottom-up network generators. In a sense, the system automatically looks >> for and validates theories on how a given network was formed. >> >> Cheers, >> Telmo. >> >> -- >> You received this message because you are subscribed to the Google Groups >> "Everything List" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to [email protected]. >> To post to this group, send email to [email protected]. >> Visit this group at http://groups.google.com/group/everything-list. >> For more options, visit https://groups.google.com/d/optout. >> > > > > -- > Alberto. > > -- > You received this message because you are subscribed to the Google Groups > "Everything List" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at http://groups.google.com/group/everything-list. > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/everything-list. For more options, visit https://groups.google.com/d/optout.
