Complex Systems Summer School 2013-Projects & Working Groups

Link title

Complex Systems Summer School 2013 Home Research Assistant Position Details Surveys/Evaluations Logistics Faculty Participants Schedule Lecture Slides Background Readings Software Projects & Working Groups CIB Markov Comedy and Tragedy in Shakespeare Global Emergent Risk NetAttac Cities Matrix Matters Synchrony on Networks Unfolding History Tutorials Blog Glue Session Forum Project Presentations Final Papers After Hours Hiking Photos TShirts Question Of the Week Page

Recursive Agents

A common theme that I've seen in complexity is that you can represent many different systems (cells, agents, civilizations, for example) with more or less the same rules. I'm wondering if anyone would be interested in modeling this? Essentially we develop a recursive rule set: agents' behavior are based on a set of their 'inner agents,' which also follow similar rules.

-Todd

Global Emergent Risk

In the current phase of globalization, networks in logistics, travel, finance, disease, energy, and ecology are growingly increasingly interconnected on a global scale. Hence, situations arise where mortgage defaults in the US bring down the Icelandic pension system, and a volcano in Iceland disrupts global air traffic. Given the scale of interconnectedness, can complexity science help develop a sense of how small perturbations in one global network might cascade into large unforeseen failures in another network? I'm wondering if we can help build a more generic grammar for describing interconnected networks and the risks posed by such systems. Would it be possible to mitigate risk at smaller local levels of scale, or do small corrections of local risk inevitably cascade into larger risks that we lack the ability to respond to (think wildfires in the Southwest where more frequent small fires would help reduce the chance of large scale fires that cannot be controlled.) I'd love to talk with anyone interested in ecology, globalization, risk, and network modelling.

My advisor at Princeton is already funded for a three year interdisciplinary symposium on Global Emergent Risk, so if people are interested, we have resources to carry on a larger research engagement.

-- Manish

- I don't know if large risk cascades are inevitable, but these are timely questions. I have colleagues at the Global Catastrophic Risk Institute who are also interested in this topic. I'm not sure what building a "more generic grammar" means, but the cross-impact balance method is pretty generic. Hopefully you can make our tutorial! Our group is kicking around methods for a project at the moment and have not decided on research questions, so perhaps we can join forces. -- Vanessa

- To clarify, what I meant by "generic grammar" is some way to describe a set of networks and cross network dependencies that can be decoupled from the content of the network. We could use this language to describe interconnections between hypothetical logistics, travel, financial networks for example, and then use simulations to understand the conditions where risks might be amplified and/or jump across from one network to another. -- Manish

- I'd love to join discussions on this as I agree that there's a need to start thinking about network of networks and not just networks in isolation. The paper by Gao, Stanley et al. on the robustness of a network of networks might be useful. -- Cheryl

Alcohol Consumption and Language Fluency

(This is just one of my random thoughts, don't take it too seriously. I'm bringing it up in case it would actually add interesting noises to anyone's thought process)
There's a interesting thing about me and some of my friends who speak a foreign language. There seems to be a kind of non-linear curve that depicts my blood alcohol concentration and the fluency of my speaking English (Chinese is my native language). For example, usually after one beer, I start to talk faster, and semantically relevant words pop up in my mind more fluently or spontaneously. Or say, I express the same idea with less stuttering and in shorter time. But after some point, i.e. 32 oz of beer, even it might still sound fluent, regarding the physical property of the speech, the grammatical structure of the sentences start to break down. Ultimately, i.e. half bottle of rum, I only repeat 2-3 very short sentences/phrases independent of the presence or the identity of my audience(s) (i.e. murmuring in English for hours to a group of Chinese friends).
We know that alcohol can influence our behavior by, among other things, binding to certain neural receptors (like GABAa?). I just thought it might be interesting to model how alcohol intake influences the dynamics of language production via regulating neuron signaling. I'm also wondering about what it might imply about our brain dynamics when we're silent.
Welcome to talk to me if you find this interesting, or silly, or you simply want to correct my grammar. Thanks!
-Mengsen Zhang

@Mengsen I'd be very interested in doing some empirical work on this topic. -Max Kleiman-Weiner

I've experienced this same thing with Spanish, Bahasa Indonesia, and playing billiards/pool. Creative topic with solid scientific underpinnings that lend themselves nicely to complex systems modeling, I suspect! – Kristen Honey

How the ocean can help us heal complex chronic disease

The human body is its own ecosystem -- much like the ocean -- with resilience, hysteresis, synergistic properties, and multi-system dynamics that depend on matrix conditions. We can use our understanding of the ocean and ecosystems to help us heal? Can we model different scales -- an ocean, a human body, and a microbial community -- to explore ecosystem/human/microbial health in a holistic context that depends on shared key elements like flow & trace minerals as building blocks for function? As one example to think about, chronic illnesses like HIV/AIDS or Lyme disease disrupt the immune system and human body functions (e.g., methylation pathways, detox pathways), preventing optimal function that weakens the human body and makes it vulnerable to other infections. It’s a downward spiral of negative feedbacks, analogous to a backed-up, atrophying ocean or estuary that causes fish kills, destroys coral reefs, etc... analogous to microbial communities that shift when environmental/matrix condition change. I suspect our knowledge of the ocean & large ecosystems, which we can see and visualize, can inform new thinking about system dynamics for health & recovery at the scale of a human body and at the scale of individual microbes & microbial communities...

Anyone else interested? – Kristen Honey

- This sounds like a really nice idea. It would be interesting to understand how the disease-mediated degradation of immune/metabolic networks (the loss or alteration of edges and nodes?) affects the response of these networks to further perturbations (e.g. asymptotic stability and resilience, transient reactivity, cascading effects of node loss, etc.). I have little knowledge of the medical literature, but I am experienced in ecological network analysis including information theoretic analyses of weighted food webs. -- Ashkaan

- I liked the idea very much. In fact I am looking at similar issues related to management of ecosystems which display hysteresis and regime shifts. Here is some pertinet information related to shallow lakes and its complex dynamics. Your idea of scaling down to human level is interesting -- Vishwa

== Join us for project discussion & brainstorming ideas, Wednesday June 5th dinner at 5:00. Everyone welcome! ==

GDELT

I'd love to play around with the new Global Data on Events, Location and Tone (GDELT) dataset, which has 200+ million timestamped and geocoded political events. Here's a writeup of it in Foreign Policy -- David

Research Network Formation

I'd be interested in collecting some data from CSSS attendants. Some kind of way to study social network formation. -- Todd

Perhaps we could collect survey questions people might be interested in looking at in a Google Doc? -- Molly

My new crazy idea, inspired by these guys, is doing something with computer vision. Maybe there's a way to photograph sitting arrangements and extract data from that? -- David

Project meeting after the lab tonight for all those interested.

Self-consistent networks for socio-economic institutions

Pablo and I started to discuss a project where we could use cross-impact balances (CIB) to investigate the implications of alternative hypotheses for interrelationships between various socio-economic/political factors. We began discussing this from the perspective of testing competing political economic theories to see what types of institutions (e.g. styles and stability of governance) would be self-consistent according to the theories. However, I would be open to other topics, including research questions inspired by GDELT. If there is interest to learn more about the CIB technique, I could put together a tutorial. -- Vanessa

There's a paper written by Brian Arthur here at the SFI that might help us frame our topic. Its called Complexity Economics. Basically sets the "rules" for thinking about economics in a whole different way. It is a very good starting point so we don't go any further wasting time taking into account economic models that are vague, non-accurate and out of date. Vanessa.

Elaborating more on the idea. If we focus on human action as the essence of culture. Understanding human action as the use of beliefs, attitudes and resources (which are scarce) pursuing a state of higher satisfaction. And culture as the sum of all beliefs, attitudes and unintentional consequences of the human action. We can state that the emergence of socio-econmic/political factors are the unintentional consequence of intentional individual actions that at the same time affect the way people act, in what they believe and what attitude they'll have toward the satisfaction of their needs. Some "institutions" will emerge as a consequence of human action but not human design. That is one of the characteristics of a "good" type of institution. (e.g. Money, language, private property, contracts, a certain type of government) and it would be interesting to test the robustness of that spontaneously emerged institution against human designed institution who's robustness is just that is law enforced. I dunno If you get my point? Compare the qualities of spontaneously emerged institutions against human designed institutions. Vanessa.?

== A tutorial on cross-impact balances and Markov chains is scheduled for Wed. June 5 at ~4:15. Everyone is welcome, even if your project interests lie elsewhere. ==

Genetic algorithms to evaluate network formation or real-world data

I have an ill-defined, wacky idea to possibly use genetic algorithms to evaluate the formation of networks as either following preferential attachment or homophily (aka similarity) rules. This short Nature paper looks at the debate between preferential attachment and similarity/homophily dynamics. I don't have a clear idea of what this would look like, but I think it might be fun to think about ways to use genetic algorithms to solve network problems. Talk to me if you think this remotely interesting and we can evolve an idea together? -- Molly

Another possibility would be using genetic algorithms or attachment algorithms to compare to models of real-world data to understand how these networks likely formed and predict future edges.

Molly, I would like to investigate using GA to create an organizational network structure (think org chart or military chain of command or even project groups at the CSSS) and then compare it to existing structures. Let's talk! -John L

Caribou Management Dynamics

This project would model caribou management dynamics in a prototype NW Alaska community during a caribou shortage. Agents in the model would be informed by data from household subsistence surveys and from management history. The goal would be to evaluate the abilities of different management strategies to achieve biological harvest goals while maximizing economic efficiencies in the community. This is a real-world problem with near-term applications. Caribou cycle on 30-to-50 year periods. The Western Arctic Caribou Herd is currently in decline. During the last caribou “crash” in this region, the state management system attempted to reorganize caribou production, which generated considerable political and social disruption, precipitated widespread passive resistance among Native peoples, and left a legacy of contempt for both management (among some Inuit) and for Inuit hunters (among some sport users). The hope is to reduce conflicts during the expected nadir of the population. Comments and cooperators welcome! Jim

Neat topic with data, I assume! I'm interested & would love to talk more. – Kristen Honey

Evolving synchronized flashes in fireflies

I was thinking about how some, but not all, species of fireflies can synchronize their flashes, as was mentioned in both lectures today (June 4). The mechanism is fairly simple, it seems, so we should be able to evolve it using a simple genetic algorithm, right? This is only half-baked at the moment, and I haven't checked to see if it has been done already, but I thought it would be neat to explore the space around these biological phenomena. More of a fun project than a serious "lets publish this!" type of project. Bonus points if we can work some neural network stuff into it. Bryn_Gaertner

How do historic facts collapse into written history?

Let's begin with a nice example: Gilgamesh, the fifth king of Uruk, decided to gather together some stories that local tribes and surrounding cultures had been telling for years, along with things that previous kings had done. This became the Epic of Gilgamesh. Someone later does a cover of the original book with some new contributions and turns it into what nowadays is the bible and the torah. Another remake of the tale turns these books into the quran, and until today... you know the rest of the story already. Peer reviewed quality, just like Nature or PNAS.

It might be interesting to study how history goes from facts to a written, definitive form which is not (and maybe cannot be) completely faithfully to the actual events. There is huge room to use, for example, models of agents that contribute to form a History with pieces of information that sums up, sometimes with contradicting versions, sometimes with hidden interests, etc etc. Furthermore, we have a great tool in the wikipedia!! We can track, for example, how many changes are made on different entries over time. We can check whether there are some generalities, how the number of edits depends on the time gone after the historic event, maybe we can quantify how successive stories differ from each other and whether there are turning points that dramatically change the whole thing...

So this is the general framework. I think this is a very exciting topic and I'd be glad to talk about this with anyone!! Just contact me! -- Luíño

Meta Food Webs

I'd like to throw out an idea I've had for a while: Most animals use space in very important ways -- predators encounter and consume prey in both 2D and 3D environments, birds and fish migrate across continents in search of resources and mates, and plant pollinators fly or walk from flower to flower, in turn providing an indispensable economic service to humans. The study of food webs attempts to understand how networks of species that eat each other persist in the face of (sometimes constant) external perturbations. Yet, network-level food web studies seldom address the dynamics of animal movement, and I see this as a fundamental shortcoming in our understanding of nature. Recently, scientists in fields like computer science, physics and neurobiology have begun to model and explore multi-level or multiplex networks -- networks of nested networks. This seems like a reasonable candidate for the theoretical study of multiple food webs that are linked by spatial networks of animal movement. One preliminary question that comes to mind: How do the number of "mobile" species and the "speed of movement" alter important dynamical properties of complex food webs at larger spatial scales (i.e. at the meta-food web scale)? I am not dead set on answering this question, and I look forward to gaining insight from scientists who study other types of networks. I'm also not set on the multiplex network framework. Potential alternatives that come to mind are IBMs, PDEs on graphs or integrodifference equations. I look forward to any suggestions or bright ideas! -- Ashkaan

Very cool topic. Definitely interested & would love to talk more. – Kristen Honey

(Evolution of) Aging

Sorry that this one is a bit long. I'd like to brainstorm with anybody interested to see if there could be a viable project in the following direction.

Let me first define aging: deterioration that happens as an organism, e.g. a human being, gets physiologically older, eventually leading to increased mortality and/or decreased fecundity.

Some background into the 'classic' theory: Evolution is about getting to be there in the future, that is, you and/or related organisms, for instance offspring. Increased mortality and decreased fecundity as such are clearly unfavorable to getting to be there in the future. Then why could it evolve? Well, evolution tends to become less sensitive to anything happening to an organism as time progresses, because events that take place at some point in time can affect only events that are future to that event. All offspring that an organism already has at some point cannot be affected anymore, and this is a non-decreasing function of time. This can be formalized, and I'd be happy to write down the math if anybody wonders.

It has, however, limited value to theorize too much about age-specific events, while in fact events at different ages are tied together in pathways of causality, dynamic change and so on; age per se is not a cause of anything, and changes at some age do not happen independent of changes at other ages. There are a number of sufficient arguments why the 'age-specific' picture does not capture this reality. I'm skipping these arguments for now, but ask if you are interested.

We have two things that matter for the evolution of aging. 1. The declining sensitivity of fitness to age-specific changes (of some standardized magnitude). 2. The fact that there are constraints that make that what happens at age x is not independent from what happens at the ages in its neighborhood. Thus, the trick is to figure out what the constraints are - this is where complexity may come in - and to combine these with the effect on fitness that age-related change has. The effect on fitness may be solved analytically, but there are various reasons why computation may be preferable, specifically the not always realistic assumptions that are necessary to allow for analytical solutions.

Mechanistically, people tend to think about aging in two ways. The first is that aging is caused by the accumulation of damage. If this damage were all repaired, aging would not occur. People then try to think of reasons why repair would be imperfect. The second way to think about it is as a gradual loss of robustness/control, sometimes in the context of reliability engineering (is anybody familiar with that?). Beautiful medical example: old people need more insulin to process a standardized dose of sugar, and their regulation shows more peaks, especially upward peaks, than that of young people, who tend to need less insulin to process the standardized dose of sugar. Of course, if you consider loss of control as a type of damage, the two are the same, but the distinction is perhaps helpful because the way people tend to think about damage is not in a dynamical way, but just as protein aggregates sitting in the brain, inhibiting the function, cartilage that looses its suspension, etcetera. Again of course, both may influence each other.

Now toward a project proposal. If I think about how complexity may (in part) determine physiological constraints, I think of the following. Usually people tend to think of repair being limited by available energy. But to repair something, the body needs to have available somewhere the information necessary to restore the initial state, and use that information at the place where the damage has occurred to be able to repair. This is where I hypothesize complexity comes in. The necessity of different components of the body to interact may put constraints on repair other than just energetic, it is also a question of whether the energy can actually be used for the repair (flow through the system in the necessary way). That may require space, a certain chemical environment, hormonal setting etcetera that may be incompatible with the proper function of an organism. (Can you repair a car when it is driving?) Also in the 'control/robustness thinking', you have to get back to the original situation to avoid aging. Is that compatible with the best evolutionary outcome?

These are just some ideas I'm throwing at you, as you see it's not perfectly fleshed out yet, which is good, because it should benefit from your perspective. So anybody interested, please let me know!

Cheers, Maarten

Neat ideas & I welcome more discussion around these topics. Aging is cool from the academic science side, if not from the personal experience side, lol. It seems to me that these same models/concepts for aging also apply to loss of functionality from chronic illness... illness expedites aging? – Kristen Honey

Quantifying Synchrony in Dynamics Occuring on Networks

Recent work has focused on developing information theoretic measures for quantifying directed information transfer, with particular applications to social media. These metrics are motivated by the work being done in theoretical / computational neuroscience on the analysis of spike trains. To do this analysis, the behavior of users on a social media platform like Twitter are treated as point processes, where we only keep track of when a tweet occurs, and ignore its content. That is, we treat a user's behavior over time as 'spikes.' Despite the simplicity of the approach, it was found to be successful in identifying key actors within real social networks.

I am interested in applying a similar methodology, but using a different measure of synchrony motivated by computational mechanics. This method seeks to learn the hidden states that generate a user's behavior (very much in the flavor of a Hidden Markov Model, but with a few key twists), and then considers the mutual information between the state sequences of the two users.

A first step for this project would be implementing the methodology proposed by Shalizi, et al., on the toy model proposed by Steeg, et al.

If that is successful (and completed quickly), I have a data set (network connections and behavior) of fifteen thousand Twitter users collected over a three month period. We are interested in using this approach to identify dynamical communities (not only users who are connected, but users who behave in synchrony) within the social network. This takes us beyond typical structural community detection that has had great success in the past decade.

These approaches should work with any sort of dynamics occurring on top of a network-type structure, so if you have a different system you would like to use as a test case, I would be very interested to hear about it!

Dave Darmon

A spin off?

Sorry for editing in your proposal, Dave, but I wanted to comment an idea I have been interested in for a time now. There is this great technique used in neuroscience to pin down the most effective time-course excitation that a neuron can get so that it fires. It basically averages the input a neuron had been getting before each of its spikes. You can find a thorough description in the very popular Dayan & Abbott book on neuroscience. Since you mention the abstraction from tweets to spikes, I would be very interesting in applying such neuro-inspired analysis to this social interactions. Neuroscience has got many more techniques, so I do not pose it as a closed matter. Just open for discussion, but seems like everybody is sleeping by now ;) -- Luíño

You're refering to the spike-triggered average? I hadn't thought of that!

All of the work I've done has only considered a single user's time series for prediction. I would certainly be interested in looking at how including the 'inputs' to the user (or at least the inputs that occur on Twitter) impact this process. The spike-triggered average seems like a great first start. Computational mechanics also has some tricks in its toolbox that could be used for this sort of input-output problem. -- Dave D

Energy resources supply patterns from biological systems to humans

My idea is to start a brainstorming, it is nothing more than some disjoints thoughts!!! And many questions without an answer so far!! :-)

I would like to investigate how biological systems obtain their energy requirements. Are there patterns or network structure that evolution has developed and that are efficient for animals/plants? Can we replicate these structures/networks/patterns in the way we (humans) obtain our energy? Can we learn something from the evolution of the energy provisioning of other species? Is that feasible? My primary idea (given my research bias) is to then apply the findings to the structure of the electrical system. Is the current centralized generation and long distance distribution something that appears in nature? Is there a more efficient way that emerged from evolution in biological context that we can use for future provisioning?

We can apply this not only to electrical systems but more in general to the way we use our resources.

Anyone interested with ideas, feedbacks, thoughts? – Andrea

Ecological networks are, indeed, finely structured both topologically and energetically. I'd very much like to talk to you about this idea in more detail. -- Ashkaan

Definitely interested & would love to talk more. – Kristen Honey

Cellular morphogenesis - The evolution of organisms' shape

I am interested in genetic design, not as much the modifications of already existing plants and animals but rather the capabilities of from-the-ground-up design of completely new organisms. One of the most basic question when it comes to multicellular organisms is how they end up with their particular shape and how it is a product of cell growth, membrane adhesion, chemical signalling etc. I suggest a project where we explore what types of shapes can emerge in simple cell growth models and how the shape can be controlled by tuning the organisms genes, the interactions. When we have something running, one interesting continuation would be to apply genetic algorithms on the growth parameters and se what happens with the emerging shape if we for example assign high fitness to high surface area but low volume, high moment of inertia or maybe concentric shells of different cell types and see what types of shapes emerge when under influence of external forces. We can also compare our results with simple, real organisms and se if nature found similar solutions.

There are different cellular models we could use, most reasonable would be a cellular potts model (B) or a vertex dynamic model (D) [1], depending on what type of details we want to include. I suggest we start out with 2D simulations.

I also believe this project can steer into many different directions, so if you like parts of the premise and have ideas on other directions we could take this, say hi.

Max Planck instutute with a group on this subject: [2]

A vertex dynamics model investigating how a specific cell morphogenesis could occur: [3]

Oskar email me at oskarxvi at gmail dot com

How big can a city be?

West et al. have discovered striking—and universal—patterns in the way cities scale with size (see West's Ted talk, short Nature paper, longer PNAS paper). Do these scaling laws allow us to predict how big a city can be?

Here's a motivating analogy. The mass of an animal grows scales the cube of its size L, but the cross-sectional area of its leg bones scales only like the square of L. This implies that bigger animals must have bulkier leg bones in order to sustain their own weight. (Think of the legs of a mouse versus the legs of an elephant.) Since the bones can never get bigger than the animal itself, this immediately tells you that land animals cannot be arbitrarily large: they must have a maximum size. If you plug in the numbers and estimate this maximum size, you find a value consistent with the largest known dinosaurs. In fact, with similar reasonings—which were discovered by Galileo, by the way—you can easily find how tall trees can be, how high animals can jump, etc [4].

Now, to run a similar argument for cities, we should understand what constraints would limit their size (the equivalent of "the legs of an animal can never be larger than the animal itself"). These constraints may be technical, social—I'm not sure. (Crimes are perhaps an example. West et al. show that the number of crimes committed in a city grows faster than the number of inhabitants. Clearly then, at some point the likelihood to get shot the next day will get too high, and people will start leaving the city.) I wonder if the social scientists among us have any insight about such constraints, and whether we can actually come up with a prediction for the maximal size of a city based on them. – Matteo

- Scaling patterns for cities are fascinating, but a potential data limitation to the PNAS paper is that the study was applied to cities in the US, EU and China. It can be argued that these economies have particular similarities that may not be transferable to cities in developing countries that are not China (e.g. Jakarta, Delhi, Manila). To complicate matters a bit more, there is little consensus on what a city is -- is it defined by the political boundary? What about the economic boundary determined by bedroom communities (suburbs)? If one takes the latter view of a metropolitan area, the population densities of some "cities" in developing countries is truly astounding. The New York metropolitan area can be interpreted as spanning 4500 sq. miles with 20 million people (a density of ~4600 people/sq. mile). Under the same interpretation, Jakarta spans only 1075 sq. miles with 25 million people (a density of ~25,000 people/sq. mile!!!). The point of my comment is that I wonder how well the scaling findings of West et al. hold up for cities in developing countries (that do not include China). It seems possible that there are scaling patterns there as well, but they might be different. By the way, I got my numbers for population densities from Demographia. -- Vanessa

- Because of national differences, there's certainly no consensus yet on how to define the boundaries of a city. Even UN recognizes this and suggests following the boundaries established by individual countries. On the point of whether scaling patterns will be the same for developed and developing countries, the scaling patterns might still be the same (wealth creation leads to superlinear scaling while economies of scale results to sublinear scaling) but the actual value of the scaling exponents might vary. It would be interesting to see whether such two sets of exponents exist as it might explain why the experience of living in two cities of the same density can be different (case in point Manila with a pop'n density of 21.9 M vs Shanghai or NY with 20.9 M). To quantify the living conditions/living experience of cities, we can use the Economist Intelligence Unit’s 2012 Global Liveability Survey. If data is now available for developing countries, it wouldn't be hard to check this. But having lived in Manila and seeing how inefficient collection of census data in such cities can be, I doubt if we can have a substantial sample of developing countries with complete data. We can probably check here.

As for setting the thresholds in computing for the maximal city size, we can use the parameter values for Melbourne as according to the EIU survey, it is the most livable city in the world. If somehow we decide that using this is not a good way of defining the thresholds and there's no other way of setting them, jumping off from Vanessa's point, what we can also do is have several sets of parameter constraints and say that if city A falls under category A with parameter constraints {A}, then this is the maximal size it can have. We can do some clustering analysis of the parameter space to determine the city categories. I think it would be worthwhile to talk to someone from the cities group here in SFI as they've already mentioned before how population growth behaves with the different scaling patterns.-- Cheryl