Welcome, everybody. Our guest spot for this unit is Professor Mark Newman. Mark is Paul Dirac Collegiate Professor of Physics and Complex Systems at the University of Michigan. He is also an external professor at the Santa Fe Institute. Mark has worked on many areas of complex systems, but most notably he has been a pioneer in the study of networks. His recent book, called Networks: An Introduction was published recently by Oxford University Press and it’s the definitive place to go if you want to learn in depth about network science. Welcome, Mark! Thank you very much, nice to be here. Our current unit is on networks. We’ve looked at some of the common notions used in network science, at least early network science, like the notion of a small world network or a scale-free network, the notion of preferential attachment. I wanted to ask you -- these were important notions in the early years. People talked about them a lot. But do you think they are still relevant? Oh, absolutely. I definitely think they’re still relevant. These are some of the foundations of the field, some of the most important phenomena. It’s true people at the research frontier are spending less time studying those things these days, but that’s because a lot of those questions have been sorted out and are quite well understood now, and that doesn’t mean that the answers aren't less important than they were. There are plenty of things that were discovered and sorted out a long time ago, relativity and quantum mechanics and so forth. Still really important things -- so I think these things are the foundations of the field. Very important ideas, but the research frontier has moved on to other things. So what’s an example of some of the things at the research frontier these days? There are many different things, but for instance I think one very important area that’s of interest to me and a lot of people these days is what happens when networks are changing over time, so most of the concepts that people discussed in the early days looking at this field were to do with static networks. How a network sits there, doesn’t change, still has lots of interesting behaviors, but when we look in the real world, most of the networks we’re talking about actually are changing on some time scale, whether it be short time scale or a long time scale, and we’re interested to know how that affects the kind of phenomena we’re interested in with networks, how we can model that, how we can understand the effects of changes One example might be the internet for example, which if you’re interested in the structure of the internet today, one day, you can pretty well treat it as a static network, it’s not changing very much, but if you want to understand how it has changed over the last ten years, well, it’s grown enormously in size and it’s grown in very specific ways. It doesn’t just grow randomly. There’s particular patterns in the way it grows, and how’s that going to affect say, the performance of the internet. Has the performance of the internet gotten worse because it’s been getting larger. How could we change it to make it perform better If we could control the way it grows, how would we like to do that? In order to give it good performance as it gets larger. These are the kinds of questions that one could tackle if one had a good understanding of what happens when networks change over time. I think that’s a big growth area. So there’s two ways you could think about that. One is you could think about networks getting bigger, but also you could think about information being propagated over networks, those are two different notions of network dynamics. That is true. People talk about dynamics of networks and dynamics on networks. There’s also interesting dynamical phenomena that are going on networks, when the networks themselves might not be changing or not significantly changing, but something dynamical is happening on the networks, like the spread of information, the spread of data over the internet or the spread of a disease over a contact network of people. How does the flu spread from person to person, or the spread of a computer virus over a computer network, or the spread of fashions or fads over a social network, the spread of a rumor on facebook these kinds of things. These are all examples of things where the network itself is probably staying roughly constant in time, but there’s some interesting spreading process or dynamical process happening on top of it. That’s something that has also been a big area of research in the last few years I would say we’ve made more progress in understanding dynamics on networks, those kind of spreading processes and so forth, than we have dynamics of networks, when the network itself is changing There’s definitely a lot of open questions still in that area. So what are you working on these days? What are you excited about? That’s definitely one of the areas that I’m excited about, for instance I’m interested in questions like if we can gain some understanding of how a network changes over time, can we make predictions of how it’s going to change in the future? More generally we like like to understand how to make predictions about networks, for instance another area that I’m interested in is can I make predictions about when network data are wrong? A lot of the network data we look at, social network data, biological network data, has errors in it. It’s not all correct. And those errors can have an effect on our understanding of the system, so we’d like to be able to understand those errors better. One thing that people have asked about is can we predict which connections in a network are likely in error It says there’s a connection between these two nodes, but maybe there actually isn’t. Can I pick out nodes, connections in the network that look sort of suspicious. They don’t fit the pattern of the other connections in the network, and say that one looks like one that might have been an experimental error. The reverse of that question is can I find connections that are not there that I’d expect to be there? I see these two people are not connected, but I would have thought that they would be, that looks suspicious to me. Maybe that’s an error, or maybe it’s not an error but maybe I should suggest that those two people might like to become acquainted, maybe they could tell each other something useful. Yes, Facebook tells me that all the time. Right, indeed. Of course. So this is an area in which people have done significant work in limited specific contexts, without thinking more broadly how the same ideas could be expanded to other kinds of networks. So the classic example of this is recommender networks like on Amazon, you buy a book and it says you might be interested in this other book. How does it do that? Well, it has a network representation of who bought which books and it tries to predict which links in that network are missing but should be added. It tries to predict the ones that are most likely to be good connections in other words, the books that you’re most likely to want. So there’s a whole literature there on how you do that recommender task. If I have some subset of information about people and the books that they like, can I predict other books that other people will like? And that’s a link prediction task, so one of the things we’re looking at, and other people as well, one area of current research is how can I do this kind of link prediction task on other kinds of networks as well, so to some extent that’s a question of leveraging things people have learned in these other areas as well, this is something we see often in complex systems, that there’s somebody in one area, in this case recommender systems, who has made a pretty careful study of this, maybe we can take ideas from there and apply them in a different context. Okay. I know you’ve done some work in the past on community structure detection in networks, where you try and look for communities that might not be apparent and I was wondering if that has seen any particularly interesting applications. Yes, so that is an area that I’ve worked on, and the idea there is I have some big network, and I believe that there are clumps or communities in this network, there are groups of people who are all friends, or there’s a group of webpages that are all linked together, can I pick those clumps out? If you draw a simple picture of a network, it’s often obvious where the clumps are, but now we’re talking about very large networks with millions or even billions of nodes, you can’t draw a picture, and you certainly can’t pick the clumps out by eye, so we’re interested in developing computer methods for doing this on large-scale data. So people have been working on this for some time. There are various applications there that one might think of, for instance I might be looking at the web, and you can look at a web network and pick out clumps of nodes which tend to be pages that are all on a related topic, or are all sort of belong to the same people, maybe it’s a clump of nodes that’s all at the same company, stuff like that People have looked at these kinds of things. You can look at them in social networks, so for instance, people have looked at various social networks and have tried to pick out the groupings in them. If it’s a social network of friends, then you’re probably just picking out the groups of friends, but people have looked at it more broadly, for example there has been a famous study where someone looked at which jazz musicians collaborated with other jazz musicians, even picked out groupings of musicians that worked together frequently. There’s a famous concept in the literature that actually goes back several decades of so-called invisible colleges, which is when communities organize themselves into groups that are not necessarily the official groups that they’re supposed to be organized into, so a classic example of this would be in academic research. We’re supposed to be organized into physics, and computer science, and mathematics, and engineering, but you actually find that the collaborations may be along completely different lines. Perhaps the collaborations are actually a bunch of these people in physics collaborating with these mathematicians on some problem, or collaborating with these engineers and so the actual collaboration groups divide along different lines from the traditional lines of the departments that people fall into and so one can apply these kinds of community detection algorithms to networks of who’s collaborating with whom. And pick out those kinds of invisible colleges. Now I know of some practical examples where people have had success doing things kind of like that I don’t yet know successful examples where people have done exactly what I just described and succeeded in determining invisible colleges that were not traditional ones, not just traditional departmental lines, but this is certainly the kind of thing that people look at. So an example of that is people can look at that in citation data, which academic papers cite which other ones, and look for clumps in citation data of papers that are all citing each other and you can maybe pick out emerging fields. Here’s a new area of research which is not yet sort of officially recognized, but you can see it emerging in the literature. You see these clumps of papers that are all citing each other, so this is definitely something that people are very interested in. Like network science itself. I think network science is an excellent example of that. Of course it’s now quite a well established field, but ten or fifteen years ago when there were only a few people working on it I think that it would have been interesting to do the study back then and you would have seen the new field emerging in the literature and actually sort of pick it out before it became a recognized area, as it is now. One last question, so as the author of a well-known book on networks, a textbook, what is your advice to people who want to go into the field of networks? How should they start? So that’s a good question. I think there’s various ways to get into it. The way I got into it was through a fairly traditional background in theoretical physics and I think that if you can do that, then that kind of thing is a good way to start, to get a solid background in whatever your chosen field is, which might be engineering or computer science or physics or biology, and then look for applications in an area that you know about of networks. I think that there is something of a tendency in this field for people to treat networks as a hammer and go looking for nails so you know, I’m a network scientist, what can I do with what I know about networks? Whereas I really feel that it should be driven by, I have an interesting scientific question that I’d like to answer, and it turns out that networks are a good tool to allow me to answer that question, so I definitely favor having a strong background in some field that really interests you and then asking how could I answer questions in my field of interest using networks or maybe other complex system tools Alright. Okay. Well, thank you so much. This has been great. You’re welcome. Good talking to you.