July 12, 2010

On Graphing Data

A few weeks ago I sat down with Jean-Baptiste Labrune at the MIT Media Lab to talk about my recent work, The Betweeners. The conversation recorded below ranges though a variety of topics relating to the representation of data and networks in particular, and how these representations do and don't convey meaning.

JBL: Hello. Hi.

IW: Hi. So The Betweeners, is a work of photo portraiture but I like to think that it's also a computational photography piece in the sense that it uses... I wrote a piece of software, which found subjects for the portrait. So, in a sense the image you see can't be divorced from a software process. So in some ways it is a work of digital art, even though the photographic process uses a very old 1950s camera and a chemical film process.

JBL: And there is also image editing...

IW: Right, and image editing and processing was part of the final print. So I used PhotoShop to assemble...

JBL: To compose.

IW: Exactly. To compose. So it was really intentionally going back and forth between using old media and new media and trying to create something that I think is symbolic of how people relate now, more than before, because it piggy-back on online social systems. And how the role of the algorithm plays a significant part in who this group is and what they are all about.

JBL: Yes. So I start, maybe, if I can interrupt you here. I would like to talk about, to reflect already on this idea of the algorithm that selects the people and start to criticize this algorithm itself. I was reading the print-out on your wall that says "edge = link, vertex = node" — this idea of graph theory, SNA (social network analysis), the analizing of centrality. "Betweeners" are linked to this idea of "centrality." It's part of the vocabulary of the graph theory that social network analysis took of representation. It's a model that represents entities that are connected. So connected entities. So it's a topological concept. It relates to what defines a network and what is interesting is that it talks more about the representation itself than these people. For me, this algorithm that is selecting these people is actually selecting entities derived from the model itself. So I think that although they are defining the identity because they have this relationship with people who have never met — actually that have never met each other although they share maybe lots of connections. This might have been one of the criterion to select them. Like they are part of the same graph, that's for sure.

IW: Yes, they're definitely part of the same overall graph, although none of them had actually met each other...

JBL: Because they have a distance between themselves, but they all share the same characteristic: being a hub, being a strong centrality — meaning a strong connectedness to people, correct?

IW: Well, the fine definition of this kind of centrality is worth noting because it's not actually a hub...

JBL: No, it's connects two distinct parts of the graph.

IW: That's right.

JBL: They are actually more connectors between fragmented, dissociated parts of the graph.

IW: Right.

JBL: It's this remarkable visual structure you see when you see this. There's another way to refer to this graph in HCI, in Computer Science called the "spaghetti graph" because at a certain point every 2d projection of the graph (because you never see the graph, you always see one projection in 2d on the screen or on paper...)

IW: Yes.

JBL: You can operate it and see it in 3D, you can variate the representation, but you can never see the whole graph at the same time. It is one of the properties — you are always occluding, seeing different things. One way to represent the graph is by clustering some parts of the connected points, and when you see this cluster some times, you see three clusters connected by a certain point. These points are The Betweeners. This point of centrality, because they are connecting different parts of the graph. Basically it could be like a hub, in a way is connecting also places for planes, you know it's like this junction where everybody has to pass through... Basically it's this: it's this point in a route where if you want to go from one point of the graph to a totally different one, you have to go through this one.

IW: It's like a router on the internet.

JBL: It's an obligatory point of passage. It's this place, where if we cut the graph in this place, then we create many graphs. The graph will not exist anymore as one entity. There is not another part that might link it, to save it, to be still connected.

IW: But the algorithm does measure this in a continuum. But you're right, one of the early ways of calculating the algorithm made use of a technique of removing the nodes in question and measuring the change. So the more central you are, the more the graph will change when you're removed.

JBL: So I'm talking about that because this centrality property is central for me in your research because it is what defines and criticizes the classic selection process of having the casting of a usual art piece where you recruit, enroll people for an artwork. Here it is a machine, an algorithmic process that helped you make the decision, because of course it's not the machine that contacted these people, it's your voice maybe with your cellphone. If the algorithm is selecting 20 people, maybe you decided arbitrarily 6 of them because its easier that they are in Montreal and not Vancouver or whatever — they are now available or not. It's cognitive help. It's a process that helps you as a person to take a decision, so we are really talking about this coupling between a machine and a human doing something together. It's very interesting in this project, for me.

So another thing, more free association, this is my we're having a free conversation with words because its faster than writing things in text for me. This is why I asked you to do that. "Nomos", the rule, the regulated. Deleuze talks about "strident," the things that has stress, the relief, the discrete, the area of the non-continuous, in a way — compared to the "anomos", the analog, what is part of this idea of a seamless surface, "sans-couture", without seams, this thing that has no discriminant. So how to go from this hierarchical property of a surface that is either segmentable, segmented, finite to a more analog thing in the progression of a graph? I think what is interesting is this central point is remarkable, it is a manifest point of differentiation. It is a nomic point, in the sense that it creates significant difference between parts of the network. Why are you interested in these people and not in randomly choosing somebody from the graph, or randomly choosing somebody from a non-central part of the graph? It's because these people are remarkable.

I would like to slow down a little bit the conversation on this idea of defines something remarkable that you might remark. I think that the remarkable in this is what is manifested by the algorithm, the model itself. I think it's remarkable in the epistimological frame of the model itself. I think that the question, and I could talk more with you, and I will in the future, about what are the conditions that let you decide to use this graph or this model of a network. Maybe it's a pure semiotic similarity between the network of people and there are networks between computers and I would like to explore this relationship between these tools we use: Facebook, LinkedIn, etc. I think that the human connectedness, again the network of people is a representation of what it is to be a human. This representation if really really far from what it is to be a human, experiencing and encountering with another human. So it's an analog.

IW: Right. It's an analog and it's a real pale comparison to I guess the rich realities of human interaction. In a way, it's a kind of a symbolic simplification of relationships, and this is something that Niko and I talked about for a while. It's an abstraction — an abstraction of this notion of friendship. There's a definition somehow where if you meet someone for ten minutes they qualify as a new Facebook contact, but it's a pale comparison to the more traditional definition of friendship or this... Well, I say "traditional", but before Facebook I found that for me (and it varies for different people) friendship is something that gradually build over time and it is somewhat of a significant label. You don't call everyone "friends." Some people are acquaintances, that sort of thing. So yes, the network graph is a simplification and its a model, an analog, but it had this tremendous blowback — a retroactive effect on people and their definitions of friendship. So it's a significant social force. It's not just an analog, it's not just a symbol.

JBL: It's not just a representation. There is an existing set of relationships between people and we might represent it in this way. It's also shaping the way people might now find their human relationships. It's also a bit like in operant conditioning. Like this idea that it's reciprocity. But it's more this idea of shaping. The idea that things go both ways. That the model that you create is a way to understand what you do, but it might also be an a priori definition of what you might do in the future. The model explains but also defines and persuades you.

IW: Explains and effects.

JBL: Exactly. The map is not the territory, but the map defines the artificial — it redefines the territory.

IW: It produces as much as it...

JBL: Not entirely. It produces, according to me, an artificial addition to the territory. Especially because the territory is not entirely controlled by the humans that created the map. I'm not sure that trees are mapping, but they still grow. But there is definitely this two-way process. Where the abstraction is elevating on top of reality and now is redefining, structuring reality. It's a double movement. It's not only a posture to observe, it's a little bit like the bias of the physics researcher's microscope: that beaming an electron is not only seeing, but modifying. But the representation that is seeking the intention is also affecting. There is a bias here.

So, what is the bias of trying to use this representation when you use a graph? How can you guarantee objectivity and how much actually aren't you talking more about graphs than talking about these people? And this is the provocative point I wanted to say: how much are you celebrating a graph structure in your piece? And by actually putting an icon on top of the nodes of this graph where you could have put actually a lot of different people...

IW: Yeah, like you've said before, it's actually something that's going in multiple directions at once. Part of how I initiated the project was really looking for a way to kind of define something that I felt represents how I operate socially. I am not necessarily concerned about having a lot of friends, but I am concerned about having friends from different social spheres.

JBL: So to confront the perception you have of your own practice with maybe a theory or a model, and to see how reflective it could be for you to see "This is what I think I am at MIT or when I was in Vancouver before. What is my own social graph and how can I confront, compare maybe, oppose this perception of myself to other people." Maybe do a little bit more work on what is this thing and maybe also validate or check or maybe invalidate if I am also corresponding maybe unconsciously to this structure that pertains, pervades in my own perception and practice of meeting people. So, I think it's a salutary — it's very important distantiation to be able to contemplate the model, but you have to be able to criticize it also. Deconstruct it. And it's always a tricky part to try to objectivize the objectification... that's not the correct word... "hypostasis" is this idea of elevating to a degree an abstraction you are actually doing this necessary reduction of reality — of maybe the reality of your perception. You will also create a new kind of diminished reality that you might then ad infinitum criticize. You will create a new graph structure that is your own critique of the graph structure.

[JB looks up definition of "hypostasis."]

"Hypostasis:" an underlying reality or substance as opposed to attributes, or that which lacks substance. "Hypostatic abstraction, also known as hypostasis or subjectal abstraction, is a formal operation that takes an element of information, such as might be expressed in a proposition of the form X is Y, and conceives its information to consist in the relation between a subject and another subject, such as expressed in a proposition of the form X has Y-ness."

So it's talking about... in a way it's an essential operation to try to abstract from some kind of a perceived reality to a kind of definition of what other realities or people experiencing realities could relate to. Okay, so it's like how the graph structure, even if we have never met in our life because we experience the same kind of patterned behaviour in the relationship we have with people might now put us in common... or we could see the situation like this.

I would criticize this a lot because I think that this would be an heritage of logics. This would make us think that logicians, because they can make the graph, might tell us about reality. I think they are telling us a lot about the graph.

IW: [Chuckle.] I see. That is a bit more clear now.

JBL: But this jump from reality to an abstraction is great, it's very creative, but it doesn't doesn't inform us about what is the real. It's another representation. When I look at the sun, I'm afraid and I say "Oh look! It's the sun!" You know, I put a label. I re-present something that is here in a different space. I'm adding something... noise maybe, but I'm not changing the sun. The sun doesn't care about me.

So it's interesting for me to see how much you affect your perception of your own encounter with people and confront this networked representation you have of yourself. Maybe you're asking yourself: "How much does the fact that I'm on the internet, using all these networks, affect my real life of meeting people?" For example. "How much am I normal, or not?" or "Where am I in the graph?" and "Is it important that I situate myself in a graph?" "Is the society I'm living in influenced by formalism? By this kind of operation?"

Maybe as subjects we can escape formalization of exchange. Maybe we are not just actors in a system? This is one way to see each other, but there are lots of other ways.

IW: I don't doubt that and I think that it's certainly true that the logic underlying the piece may not be that sound. And certainly network graphs may tell you something about a hierarchy of importance or a hierarchy of wealth of nodes in a network, but it's also not a highly dynamic model that I was using. It's not like...

JBL: It's a snapshot.

IW: It's a snapshot of a significantly large chunk of time. It's basically the current state of the Montreal MySpace network, from when it started until now. So it builds all these connections between people, but there is no quality to the connection where I incorporate knowledge of what that connection means. It could have been someone that someone worked with five years ago and hasn't talked with since...

JBL: And even if you would know that, it will never be enough because, I would argue, that even to ourselves or to the people who might help you describe the quality of this connection... imagine if you could meet all these people and they would tell you about every person they meet, they would miss lots of details. A big part of what we exchange when we meet a new person escapes from us. Again, it's such a reduction...

IW: Yeah, but I don't think that the opposite, which is this kind of a quest for a totalizing approach to understanding people is worth pursuing either. I think it's much more interesting to take some kind of aspect and explore that, or like I was doing, which was conflating it. Taking this notion, which seems to have become very trendy over the last ten years, which is network science and network graphs and take what is typically a geometric diagram and translate it into something which is more of a kind of rich human portrait or representation.

JBL: I think that the visual representation of its geometry is not the best definition. I think it's more one of the ways to understand being connected, like a set of connected entities with a direction or not...

IW: Yes, but I'm thinking just in terms of how people understand what a graph is — a network graph. I think that these diagrams of dots connected with lines are very prominent representations.

JBL: I'm not saying that they are not part of the culture, but again to slow down a little bit. The difference that was very very clear for me when I worked in this computer science lab for my Ph.D. that they have a team called AVIS specializing in information visualization, the have a team on graph theory (mainly mathematicians). A graph is a definition of elements that are a part of the same ensemble, the same set. They are connected in some way and the nature of this relationship between the entities is very interesting for theorists. And then, once you define the graph, you can start to observe some of its properties. And the visual representations are infinite. It's a bit like thinking about OpenGL and 3D: there are many ways you can put the camera and you could explre the graph in almost infinite ways.

So every snapshot of this graph would be according to a specific discursive rhetoric point that the configuration of presentation of the graph is already a semiotic moment. It's already a moment of...

IW: Yes, digital information definitely has this property of: you have the model and the view. You have the data and the...

JBL: Sure. The data is immaterial. Like mathematics.

IW: You do have a convention which comes down very strongly in terms of different kind of data are represented. There are conventional visual representations...

JBL: It is not because it is conventional that it is true or meaningful. It's not because it's shared by people that it adds more...

IW: But that shared convention has meaning.

JBL: Yes, but it's a socially constructed meaning. It's not linked with the graph itself. It's not because you know something that everyone knows that it's true. We can all share a totally biased view on something.

One example that I criticize a lot here at the Media Lab (but nobody hears me because I'm adding some amount of itching power to the system) is that, you know, when you have a graph, a connected graph — a spaghetti noodle — it looks scientific. It's serious, right? Like, you know you have all these dots and you have colors and, wow, you put this on a map, you put it on a great science brand like Nature or MIT: that's serious. "This I don't understand because it's not by field, but it must mean something complex." Man, this is a complex set of points like you have a complex Sol LeWitt reenactment, where the algorithmic trigger is, you know, a simple rule that creates a complex structure. But it's not that because this structure is complex that it's linked with complex meaning, maybe you could then resynthesize to the representation itself. A lot of things I see in network science, network theory and representation, lots of what I understand is that they are again and again and again reenacting, re-presenting the properties of what it is to be a graph. Power laws, you know, this kind of verification that everybody does on this graph helps us understand sometimes not to believe that the representation is actually just recreating itself. Just a tautology. You know, "women are women, right?" What does it mean? Instead of being a circular definition of the model by itself representing itself. So the fact of finding connectedness in a connected graph — the fact of finding complexity in a big complex dataset.. Oh my god! Thank you. This is an obvious result.

IW: I think the interesting realm of science which departs from the basic element of asking a question, understanding how to answer it and then following through with an experiment. That classical view of the essence of science seems like almost old-fashioned. We have all these wonderful tools and programs which can help us churn through data and make these artifacts which look scientific...

JBL: Yes, they have scientific style. They are complex. I would define them as hypnotic. They are fascinating. In French you would say they are "ravis," like "ravisseur" — this guy that is kidnapping... it is the same word in French for "ravir" meaning to fill with joy, but also to kidnap. So you are taken from where you were to this space of the graph! And your eyes start to through these names and these nodes and these links. It's like this connectedness where, like a new kind of classic representation of a story in 2D or in 3d invites you to read, meaning to go through the path and does not really invite you to understand. You actually "overstand." Understanding would be to deconstruct the algorithmic construction of the graph. This is not sexy. This is more like logic or mathematics. "You ask me to explain, but it's complex and if you're not from my field I will not take the two months to explain to you how I could build this thing." The deconstruction of it will be very boring, like in law. You know, if you are interested in deconstructing why patents and things work, you know, if you have to spend five years in tribunal to see very piece of evidence in a case... this is boring! I don't want to spend five years. I want instant orgasm. This beautiful, colorful representation of whatever. Maybe it's true, maybe it's not. How do I know? I don't have time for that. I have other things to do in my life. I have emails waiting for me.

So this acceleration and this lack of tolerance for understanding complex explanations of how complex representations were crafted. And who has this knowledge? Experts. And then, there is this moment where we have to believe or not. The fact that you can deconstruct the properties of a graph requires knowledge. The epistemology of representing knowledge or information through graphs is a complex science that requires you to know lots of authors that are not very mainstream.

IW: This is one of the wonderful things about being an artist who works with science. You can try and explain stuff to the curators and the galleries, but at a certain point they just take your perspective as defacto truth.

JBL: Especially if you are coming from MIT!

IW: Especially if you are coming from MIT. It's very easy to say pretty much anything and it will be valued and respected.

JBL: [Working on his computer.] I'm looking for an artist in '86 that did a pseudo graph of the internet... a clear, fake and beautiful aestheticized representation of what was the internet... It reminds me of this article you may have read yesterday about the representation of geopolitical configuration with PowerPoint. It was all about PowerPoint reducing our understanding of the reality of people living in places... In the past I would have won an argument with my voice! Now, with the mediatization of trials, you have to be tanned and have white teeth. You might be on TV with a normal voice (because now we have microphones). And maybe now you need a multi-touch infoviz spaghetti noodle dynamics-changing eye-candy graph to create new justifications for your rhetoric. It's visual rhetoric.

I read a quote that in the past people that could not read text could have been a problem in the future of people who could not read images. This is this idea that we went back from a more text-based culture to a more visual culture where it emphasizes more archetypical, more gestalt perceptions and rhetoric. If you can do something that is stylish enough you will win, you will "ravir", you will take me from where I am because I want an emotion. I want to relate with you not just on what you say, because maybe it's too complex and I don't have time, and to to have something with which I can relate: to be a part of your graph.

IW: So I think you're taking the road of the cynic...

JBL: Not cynic. More a sceptic. A little bit different. And I'm not a sceptic, because I love graphs, but I love them for what they are. Not for what they are sold as.

IW: What are they?

JBL: They are the definition of relations between entities that inform us that our perspective on which we can maybe define or construct a perception of reality. But they are one of them.

IW: What are they useful for?

JBL: I think like all structures, they are necessary reductions. They help us to communicate things linked to the spaces they create. So, like language, like verbal language, for me is a necessary reduction. Of course, it is not representing the simultaneity of what happens in my body and my brain, but already it organizes sequentially the transmission of something that maybe would have been kept inside of me. It allows me to externalize something that might be a lie but that is already an operational one. It makes me relate to you and connect to you. So in a way it's a media in both senses of the term. These graph representations are a media in the sense that they connect me, they are connections, but they put us apart at a distance.

IW: So one of the ideas that I was bouncing around earlier on in this project was how the 20th century was characterized by the bell curve, in relation to where we are now, which is much more typified by power laws.

JBL: So the bell curve has a centrality also. Very interesting. From the bell curve to the power law. That could be a great name for an article.

IW: The initial image that I had for this exhibition was exactly that: an image of a bell curve and an image of the power law graph and it's amazing how the bell curve really enabled the construction of the notion of the "middle class" and now we have this space of the "long tail" and how it's basically... I hesitate to say that it's rich and poor because there seem to be very convincing statistics coming from UNESCO which shows that by and large as a whole is becoming less and less poor. Meaning that people are living longer, healthier lives, having less children and staying out of poverty, meaning: they have enough money to sustain a relatively good quality of life. And this is a general trend. So in my mind the notion of the long tail is "them" vs. "the rest of us" or it's "us" and "them," depending on where you are in the graph, but there's this massive difference in scale between the top 1% and the rest, and yet at the same time it doesn't seem to be one in which one might go immediately to say that "the world is poor." But in fact there is 1% that is incredibly wealthy an there is this long tail that is simply everything else. And everything else is more or less the same.

JBL: Again, how much is this representation connected to reality? For me this is a big topic of investigation. The lack of quality in the links... because we don't have this qualitative "anomos"... it could be infinite. For me the distance that can separate us could be infinite, in our subjectivities. It is very important to try to understand how much we can still relate when we are so different.

So it's not enough to draw the graph and say "That's it! Look! Look! We are connected. I can show you the graph. I can prove it to you."

So maybe we can finish on this because I started with a cultural studies point. I think that a very hot topic in the sixties and seventies, you know: Foucault, the French cultural studies, lots of people tried to address the critique of the hierarchy, the linearity of it: there is a chief, there are employees, there is a high class, the middle class, the low class. And they tried to replace it with the idea of this rhizomatic, networked... these networks are the children of Deleuze, by the way.

We are now in a moment where maybe we can try to understand how much they became a new rhetorical device, where people can use them to convince others of the complexity of scientificity, the serious — it's honest, you know, because Deleuze said it, right? It's a rhizomatic, blah blah blah whatever...

So, again: this idea that now you don't convince with the hierarchy. It's not the hierarchy of the Church, of a certain order. It's more the hierarchy of these "connected entities:" actors in the Actor Network Theory where now you have groups and they communicate and so what are now the new power places? What are the outliers, what do they mean? THe graph's understanding of the world informs us about the world. I believe that the graph's representation of the world informs us about the graph's representation of the world, first. And of course there are a lot of things that we could know from it, but I'm not sure what they are and I do believe that before understanding what graph representation can give us, we need to understand what is graph representation. Very much.

For example, the fact that the visual representation of a graph structure is not a graph structure. It's a visual representation of it. It's a reduction. It's already taking a part of it and making it preeminent and occluding another part of it. It's very well addressed in InfoVis literature. The way to avoid this is to have multiple representations at the same time, especially through adjacency matrices where you can have the top of the graph being represented through rows and columns and you can permit...

IW: I think, certainly the representation of information still has this legacy of the printed page which cannot change so there's still a primacy given towards illustrations which are fixed in space, which are projections of more complex data. With time this will change as more and more of our experience with media comes through dynamic systems.

JBL: Yes, first it's very interesting to see how it's very difficult to represent actually multivariate systems — dynamic data structures with graphs. You can have the same exact data structures, but every representation will infer something totally different, hence the importance of tools to variate the representation of the graph, to be able to discuss, to debate and maybe to compose and recompose. I'm making a constructionist point here. By building the graph, you might understand a part of it. And it's by confronting your understanding of the part of your problem that you do understand, with other people, that you might then have a more phenomenological understanding of our interpretation and understanding of the complex situation.

You know, in French we say "cerner." It means "going around" like "cerneaux de bois" is a piece of wood that goes around a middle-age garden, where they would put earth at the height of people so they could garden comfortably. You know these medeival gardens? They would use pleaching so the wood would grow around. So one of the phenomenological ways to criticize one truth, to have a perspectivist representation of a complex phenomenon is to multiply the angle of attack. It's a syncretic point. You want to go the summit? There are many roads to get there, and it's maybe in the diversity of the roads that we might approach an understanding, but we will never get there. It is an asymptotic quest, compared to this: "I have a graph and I know the truth and I can predict the future, right?"

So let's just finish on this. So it's very simple, you see: cyclic or acyclic graphs, represented in some time by tiling or non-tiled versions. This one is a bit more complex, so you see the complexity might be one thousand... One of the references that I love is this French cartographer called Jacques Bertin who wrote this famous book called "La Graphique," which tried to be a complete reference of all the various kinds of visual representation you can use to represent data, from the 50s and 60s. He was working in the EHESS (l'École des hautes études en sciences sociales) and he was a specialist of abstraction — how you make maps. He was a very big
inspiration for Edward Tufte.

With one of my colleagues we went to his house one day and I took pictures of one of the devices he made to do these adjacency matrices. [Shows the photos.] So one of the first things you would study would be all the bastard and legitimate children of the Kings of France. These are gigantic matrices. This is a way to put a map into a set of tables that are actually adjacency matrices. The theory here is that if you can build this table, you will remember how to remap the map. So it's an education theory based on a seven year old user. It had amazing results. They would ask children to do maps with all sorts of climates: mountainous, coastal, oceanic, etc. and then to hatch. So it's how having many steps of representation can help you understand and build.

So to answer again your question you asked me "what do you think about graphs and graph structure?" I love representations when they are operated, not just received. Not just persuasive rhetorical devices, but actually tools to build and engage actively in the construction of knowledge. So if you would do the graph with the people you take photos of, I think that would be interesting. Because maybe then you would learn about why they are central. You would start to access this quality of the links.

IW: The work as I have it now has two sides, but this is a welcome departure from Tufte as the authoritative narrator of the assessment of graphs and info visualization. I think these blogs like Information is Beautiful are all trying to his epitome of the most Tufte possible graph.

Ok. Thank you for this conversation.

JBL: Thank you.

Jean-Baptiste Labrune is a postdoctoral associate in the Tangible Media Group at the MIT Media Lab. His research aims at developing Creativity Research Tools (CRT) that allow artists and scientists to document and explore their own creative processes. He is particularly interested in Exaptive Innovation, Art & Science collaborations and the future of Playful cultures.

Before joining MIT, Jb earned a MS in computer science in Conservatoire National des Arts et Metiers (2004), and Phd degree in computer science (HCI) at Université Paris-Sud and INRIA Futurs (2007). He taught in art and design schools in Europe such as the Interaction Design Institute Ivrea, Mediamatic in Amsterdam and Les Beaux-Arts in Paris. He also taught in scientific centers such as Paris VI University, the Cité des Sciences and the Institut Pasteur.

No comments :