Complex Systems Summer School 2014-Project Presentations

From Santa Fe Institute Events Wiki

Complex Systems Summer School 2014

Use this space to post project presentations and outlines. Include group members, a brief outline, and your slides.

Presentation Sign Up Page

Presentation Notes

Towards a Unified Theory of Biodiversity [slides]

Group members:

  • what determines species diversity?
  • neutral theory (death rate, birth rate, speciation rate same across all species)
  • metabolic theory (all three assumptions are wrong, scaling laws)
  • their model: define rates using only one free variable
  • can get some qualitatively similar scaling to metabolic theory by limiting energy in system by preventing some births

Software and Living Systems: In Evolution a Software Engineer? [slides]

Group members:
Fahad, Renske, Diana, George, Ana Maria, Stojan and Ernest

  • cycle: innovation <-> (constrained by) conservation <-> efficiency
  • Hox genes (segmentation of body parts): modular control (each piece corresponds to a different body segment), order is preserved, innovation through duplication of entire cluster
  • telecom system: modular, core features are conserved, duplication used for testing etc. which can lead to innovation
  • future: how can they inform and guide us

Information Theory of the Heart [slides]

Group members:

  • model electricity flow through the heart using two models (probabilistic CA and deterministic PDE) on a 128x128 grid of heart cells
  • four wave patterns: point stimulation, spiral, anatomical re-entry, multiple wavelets
  • entropy of single cells can't differentiate
  • mutual information can identify sets of cells whose excitation times overlap
  • future work: transfer entropy
  • Q: comment: Granger causality is a more general version of transfer entropy and may be more effective for numerical rather than symbolic data

Tradeoff between Division of Labor and Robustness to Perturbations [slides]

Group members:

  • optimum depends on environment, focus on microbial ecology
  • assumptions: leaky functions, adaptive gene loss
  • "black queen": for each essential task, somebody's got to do it
  • threshold phenomenon: when disconnected, more robust; when giant connected component, could make system very vulnerable to environmental changes
  • real-world data, then agent-based simulations
  • shocking system removes high-degree nodes
  • future: explore other systems

The Architecture of an Empirical Genotype-Phenotype Mapping [slides]

Group members:

  • three goals:
    • explore internal structure of genotype network
    • explore similarities between genotype networks for different species
    • understand biologically-inspired context
  • link between two binding sites if they differ by a point or shift mutation
  • findings
    • short geodesics but large diameter
    • positive assortativity of node degree and level of mutation
    • networks studied have nearly optimal accessibility (measures reversibility of evolution)

Are Ants Urban Planners? [slides]

Group members:
Claire, Diana, Morgan, Bernardo, Michael, Alex, Ernest, Alberto, James and Rohan

  • qualitative similarities between ants foraging patterns and city transportation networks
  • questions
    • underlying mechanisms
    • different kinds of networks (ant foraging vs ant travel network between nests)
  • hypotheses:
    • simple foraging model can produce properties common to both city and ant networks
      • balance energy use with benefits from exploited resources (cost-benefit)
    • a few basic model parameters can describe some of the major qualitative differences (e.g. bottom-up vs top-down)
  • simulations using two models (agent-based and global) on multiple scenarios (e.g. lake, river)
  • future: explore impact of varied terrain, what is the role of central planning
  • Q: can this be used to suggest likely archaeological sites to explore next?

Structural and Functional Robustness in Networks [slides]

Group members:

  • ability to retain function in the presence of variation
  • three models: sandpile, road network, social organization
    • sandpile: probability of cascade failure
    • road model: two road types, extra cost to switch; effect of removal on simplest path or shortest path
    • social org: "failure" is reduction of performance rather than node removal
  • future: consider effect of relational cohesion

A Complex Fishery System in Northern Cameroon [slides]

Group members:

  • complex social-ecological system; messy, scarse, quantitative and qualitative data
  • focal points:
    • understand dynamics of canal proliferation
    • understand decision-making process
  • approaches:
    • game theory: multiple models and perspectives; limited by Folk Theorem, abstraction of mechanisms
    • agent-based: look at wealth vs number of canals; limited by ad-hoc assumptions and params
    • maximum entropy: ; limited by lack of understanding of mechanism

The Evolution of Inter-disciplinary Fields [slides]

Group members:

  • high risk, high reward? examples of inter-disciplinary work not doing as well or hurting careers
  • instead of looking at success of i-d vs non papers, look at i-d vs non researchers
  • by combining APS with arXiv data, can analyze whether prob of or time to publication is different between i-d and non work
  • are more authors engaging in i-d work? which fields are more or less i-d over time?
  • entropy of subject areas of work as measure of inter-disciplinarity
  • lower entropy is correlated with higher eigenvector centrality
  • future: study dynamic network of author and/or field connections

Quarantine Strategies in Financial Networks [slides]

Group members:

  • banks becoming more centralized, network more connected; too big to fail or too central to fail?
  • traditional lit: risk-sharing, robust-yet-fragile, resilience as function of topology
  • their approach: containment strategies for shocks (minimize post-shock loss to system), inspired by epidemiology
  • simulations: system regulator (recapitalize), self-quarantine (cut edges), and both
  • results:
    • given self-quarantine, additional benefit of regulation is small
    • system not fragile to just large banks, but to magnitude of (possibly distributed) shock overall

Topological Modeling of Power Networks [slides]

Group members:

  • questions:
    • what are reasonable generative models?
    • can these be generalized to other infrastructure networks?
    • can we infer robustness from structure alone?
  • spatial networks, cost of creating new links
  • challenge: geo-coordinates of nodes are not typically available (for security reasons), want to infer using weights and distances of links
  • use network layout algorithm, multi-dimensional scaling
  • existing generative models do not match properties of their dataset (Polish power grid)

Norm Propagation and Reinforcement Learning [slides]

Group members:

  • norms constrain behavior, decrease diversity
  • people have social identity, formed both by genetics and choice
  • change identity based on disposition, which is affected by observing others
  • two belief propagation models: ? and Bayesian
  • future: impact of external influence/information, cognitive dissonance

A Model of Revolutions [slides]

Group members:

  • a revolution is not just a change - it's a phase transition
  • model:
    • agents (like or dislike dictator) are the nodes in two networks: government and society
    • choose random node, affects opinion of underlings in gov network
    • interaction society: change opinion based on what your neighbors think
    • in each "round," do gov op, then society op
  • studied two parameters
    • proportion of agents activated in society each round (curfew, internet, etc)
    • alpha (material gains, loyalty, religious or ideological identity)
  • time until phase shift very sensitive to initial conditions
  • questions
    • are revolutions less likely if those higher up in gov are central nodes in society?
    • study dynamics
  • Q: what defines a revolution?
  • Q: feedback that chaos often shifts public opinion towards wanting a strong leader

Cough Analysis [slides]

Group members:

  • pulmonary diseases/conditions categorized into chronic and acute cases
  • cystic fibrosis: thick sticky mucus, not like "Triscuits in your nose"
  • should be able to detect differences in wave form using ML or other techniques
  • online database already running (but has many undesired "features")
  • future: develop classification tool, handle other bodily sounds

Co-evolution of Anti-vaccination Sentiment and Flu Infections [slides]

Group members:

  • how to measure efficacy; can only measure if person went to the doctor
  • risk perception differs by geographical region, shaped by news
  • positive correlation between sentiment and vaccination rates
  • homophily and contagion favor negative sentiment
  • questions:
    • is sentiment correlated with current flu incidence? past?
    • can incidence and sentiment collectively predict coverage?
  • from Twitter data: pos/neg sentiment, coverage; lots of technical challenges
  • behavioral transitions: "never getting shot again"

Disease Spread in Multiplex Networks [slides]

Group members:

  • food webs with parasites, really changes food web structure and dynamics
  • multiplex network (spatial multiplex model):
    • trophic transmission layer (food web: eating infected prey/vector)
    • vectorial transmission layer (vectors sucking blood of infected hosts)
  • disease spreading dynamics: can spread in either layer or both
  • future:
    • investigate role of each species in each environment
    • effect of distance and spatial structure

Fractals and Scaling in Finance: A Comparison of Two Models [slides]

Group members:

  • extreme events (fat tails): common modeling assumptions do not match real data
  • Levy process
  • self-similarity across scales, memory of past, long-term behavior
  • "volatility clustering" = burstiness
  • fractal Brownian motion allows dependence of processes, but does not have fat tails
  • idea: fractal Levy process, composite of two functions
  • subordination
  • generalized Hurst exponent
  • questions:
    • how do fractal Levy process and Subordination compare when it comes to prediction?
    • are there other processes that are good models?
    • what are the economic mechanisms behind fat tails and long-dependence?

Coupling of Networks: Non-linear Effects of Pesticides on Food Web Dynamics [slides]

Group members: Hiroshi Ashikaga, Beth Lusczek, Nhat Nguyen, Sanja Selakovic, Brian Thompson

  • model two interacting networks; each has different nodes, different edges, different dynamics which can affect one another
  • focus in particular on pesticides and a bio-energetic food web
    • 8 pesticides, 17 species in soil food web
  • model:
    • bio-energetic food web: birth rate, death rate, predator-prey, intra-species competition
    • cross-network: effect of pesticide on death rate of each species, non-linear effects of multiple pesticides on each species (can be additive, synergistic, or antagonistic)
    • master equations relating populations of species and concentrations of pesticides
  • future work:
    • model consumption, decay, and inflow of pesticides in the system (e.g. degradation of pesticides through consumption by species, bioaccumulation)
    • analyze sensitivity of the food web to changes in pesticide concentrations
  • take-away message: simulations and mathematical analysis can help us understand and anticipate the consequences of pesticide use on an ecosystem

Consciousness: An Emergent Phenomenon? [slides]

Group members:

  • agreed on:
    • consciousness is a unified subjective experience
    • project should be limited to human consciousness
  • question: can we perceive the emergence of consciousness in an artificial agent?
  • integrated information theory: anything that can perceive, process/integrate, and emit information is conscious
  • their model
    • hierarchy: neurons, nervous system, brain structures, brain, organism, environment
    • integration of info, but also hierarchical anatomical level, emergence from interactions
  • can we apply AI to simulate conscious agent?
    • Minsky: don't try to capture everything in one word, break down into multiple parts (suggests 20)
    • focus on: self-reflection, decision, reaction (fits model of AI brain)
    • consciousness and HCI: they believe that all 20 functions can be simulated, interaction is essential
    • can humans recognize X in the computer? maybe not, so instead suppose we have an oracle that can
  • future: AI experiments can allow us to simulate these functions, which the integrated info theory approach can't do

Persistence of Pollination Systems [slides]

Group members:

  • many pollination systems rely heavily on synergistic relationship
  • nenctar robbers eat pollen but don't help fertilize
    • but could have indirect positive effects
  • model 4 species: open flower, tubular flower, pollinator, robber
    • ODE model
    • coexistence when pollinator does better than robber on open flowers but less on tubular, and ? > 1
    • abundance over time is chaotic? with linear center
  • agent-based model
    • analytical model must assume good spatial mixing, but spatial dynamics are essential for pollination analysis
    • model movement of bees, energy intake and consumption, reproduction
    • flowers reproduce only if pollinated by same species (open/tubular)
  • results:
    • if more pollinators than robbers, then pollinators and open flowers survive
    • if more robbers than pollinators, then everybody dies
    • space and stochasticity can have an effect
  • future:
    • sensitivity analysis
    • find parameters to make ABM match analytical model
    • allow for mixed or dynamic strategies of bees
    • out-crossing

Bitcoin [slides]

Group members:

  • focused on three major events (volatility is interesting)
  • look at network, highly connected core, long chains thought to be money launderers trying to mask trail
  • measures of complexity, order, and freedom (MacArthur and U...?)
  • real systematic change in structure of network during different periods captured by various measures
  • network properties -> ? -> bitcoin dynamics; what is the "?"
    • sentiment analysis of newspaper articles, blogs, etc.
    • observed differences in sentiment around key dates

Social Institutions and Economic Inequality [slides]

Group members:

  • model relationship between economic inequality and economic growth
    • Kuznets curve (wealth balanced, then inequality increases, then stabilizes back to equality)
    • ?: inequality induces democracy
  • existing analytical model is flawed for several reasons
  • their (agent-based) model
    • question: how does inter-class marriage affect timing or shape of Kuznets curve
    • model via assortativity coefficient
  • results: lower assortativity yields a Kuznets curve that is shorter and earlier
  • future:
    • sensitivity analysis
    • model collective action
    • look at empirical data
    • how can we inform policy with this kind of model?

Drunk Game Theory [slides]

Group members:

  • perceived payoff changes as function of individual state
  • model
    • cooperate = buy beer for oneself and partner
    • defect wait for other
    • payoff = when sober, net gain of beers (free - paid-for); when drunk, net intake
    • parameters: coefficient of intoxication, sobriety constant (per person)
  • assumptions: uniform distribution in bar; ?
  • analytical study: vector fields, fixed points:
  • simulations: fixed point reached depends on initial conditions and parameter values; matches analytical predictions
  • summary
    • new branch of game theory and mathematics!
    • in evolutionary game theory (EGT), players change; in drunk game theory (DGT), the game changes
  • future: genetic predisposition, pre-drinking behaviors, segregation vs socialization