|
Fall 2013
Titles and abstracts
Wednesday September 18th at 12:30 pm
Optimal Foraging and Multi-armed Bandits
Vaibhav Srivastava
The multi-armed bandit problem is a benchmark problem to study exploration-exploitation tradeoff in decision-making with uncertainties.
We focus on three variants of multi-armed bandit problems, namely, the standard multi-armed bandit problem, multi-armed bandit problem with transition costs, and multi-armed bandit problem on graphs.
We focus on the case of Gaussian rewards and study these problems in a Bayesian setting. We develop the upper credible limit (UCL) algorithm for the standard multi-armed bandit problem and show that it achieves a logarithmic cumulative expected regret. In the context of the multi-armed bandit problem with transition costs, and the multi-armed bandit problem on graphs, we extend the UCL algorithm to the block UCL algorithm and the graphical block UCL algorithm, respectively. We show that these algorithms also achieve a logarithmic cumulative expected regret and require a sub-logarithmic expected number of transitions among arms.
We observe that the multi-armed bandit problem with transition costs and the associated block allocation algorithm captures the key features of popular animal foraging models in literature including the Levy flight model and the ballistic relocation model. We argue that the multi-armed bandit problem is a plausible model for decision-making in animal foraging.
back to schedule
Thursday September 19th at 4:30 pm
Social communication and group synchrony in deep-diving pilot whales
Frants Havmand Jensen
Group-living animals depend on signalling mechanisms to retain contact between individuals and prevent the group from losing its spatial and social coherence. Animals inhabiting areas of limited visibility, including underwater environments, preferentially use acoustic signals for this purpose. I use pilot whales as a model organism to develop methods for studying acoustic interactions and information flow within cohesive groups of animals to understand the social dynamics of toothed whale societies. Using acoustic and movement recording tags placed on individual short-finned pilot whales, I show that pneumatic call production is severely hindered by the high ambient pressure at typical foraging depths of 600-1000m. I show that the majority of calls produced below 500m of depth are too faint to be heard by conspecifics remaining at the surface, and that animals presumably lose acoustic contact with the social group while foraging. I show how complex tonal calls may allow individuals ascending from a deep foraging dive to re-establish contact with group members following periods of spatial separation from the social group during foraging. Determining the source of an acoustic signal, especially outside of a foraging dive when whales are close together, is still problematic. To overcome this difficulty, I will describe how I now tag multiple closely associated pilot whales to study social synchronization and flow of information within a cohesive group of animals. Preliminary data from long-finned pilot whales show how these animals coordinate their foraging dives to a remarkable degree, yet limit intragroup competition for resources by spreading out spatially during foraging. These animals employ multiple types of acoustic signalling mechanisms to mediate cohesion across varying spatial scales. Further studies of the underlying processes whereby toothed whales coordinate group behaviour and interact with conspecifics will ultimately lead to insight into the social structure and decision-making of toothed whale societies.
back to schedule
Wednesday September 25th at 12:30 pm
The Stabilization of Cooperation by Discriminators Using Imperfect Information
Eleanor Brush
Reciprocity, ostracism, punishment, and reputation are some of the mechanisms that are known to stabilize cooperation. These mechanisms differ in their details, but they are similar in that some members of a group observe the behaviors of their peers and use this information to decide how to behave in the future. It is often assumed that if there are such discriminating individuals, they have perfect information, pay no costs for gathering and using information, and can observe everyone equally. Even so, the presence of discriminators is not always sufficient to prevent defectors from invading the population. Is it ever possible for discriminators to stabilize cooperation? And how does this depend on how much information the discriminators store and use? If the discriminators observe too few agents or remember too few of their observations, defection will always dominate. Surprisingly, if the discriminators increase the number of observations they make, they can actually make it easier for defectors to invade. However, if the discriminators are less likely to observe other agents than they are to observe themselves, it is possible for cooperation to be stabilized.
back to schedule
Wednesday October 2nd at 12:30 pm
Long-term heterosexual relationships and routes of HIV transmission
Jonathan Dushoff
The proportion of heterosexual HIV transmission that occurs within
stable partnerships, as opposed to among single people, or in
extra-couple relationships, is a subject of active debate, with strong
relevance for planning public-health strategies. I will discuss
ongoing efforts to use a hazard-based approach to try to disentangle
routes of transmission from publicly available, large-scale survey
data from sub-Saharan Africa.
back to schedule
Wednesday October 9th at 12:30 pm
Testing community-wide patterns of time-covariance in Neutral communities (sensu Hubbell/Kimura/Wright-Fisher) - solving an unsolved PDE!!!
Alex Washburne
I've got a beautiful PDE whose solution can help us test whether or not an entire community is behaving according to Neutral Theory, potentially allowing us to quickly & easily use observational time-series to make stronger statements about the nature of community assembly and the dynamics of coexisting organisms' relative abundances in a stationary environment. This test involves transforming the data (finding a function f(X_t) from R^n to R, where X_t is my vector of relative abundances at time t) such that the variance of the resulting transformed data will be equal to the time-elapsed between time points (this is called a "constant-volatility transformation", or CVT) as such processes with constant volatility are manageable to test with existing datasets containing many species and relatively few time points. Finding a CVT requires solving a breathtakingly handsome, highly symmetric PDE which I believe illuminates some more subtle symmetry and features of Neutral dynamics that can be exploited for future tests and inferences about the nature of stochastic community dynamics. Come to lab tea and I'll motivate the PDE and show how we solved it in 1-D and 2-D, and how we've scraped up the wall trying to solve it in 3-D (we need to solve it for n-dimensions!). This will be a very mathematical chalk talk, but I hope to make the math accessible to people who haven't been drinking the Ito Calculus + mathematical statistics punch! Get psyched!!!
back to schedule
Wednesday October 16th at 12:30 pm
The interactive roles of resources and genetic pleiotropy in producing life history tradeoffs across ontogeny
Ann Tate Thomas
Across the ontogeny of an immature organism, some physiological pathways play multiple and disparate roles in the expression of life history traits. The shared genetic architecture of development and immunity in insects is a particularly compelling example, as many immune signaling and effector components also function in development and growth related processes in uninfected individuals. Juveniles are in danger of contracting pathogens long before they emerge as reproductive adults, and the risk of infection is often substantial when they are still employing these pathway modules for developmental purposes. In this paper, we present a within-host model of resource acquisition, allocation, and antagonistic pleiotropy between development and immunity for specific immunological pathways, while accounting for the ontogenetic exposure risk of common insect pathogens and parasitoids. We find that low resource acquisition conditions and highly virulent pathogens can exacerbate the relative survival advantage of expressing strong but pleiotropic constitutive immune pathways that exert trade-offs on development time and reproductive potential in the presence and absence of infection. The dynamics of parasite or parasitoid exposure across ontogeny can enhance or mitigate the magnitude of these trade-offs. Overall, these results give insight into the interactions between resource allocation and genetic architecture of pleiotropic traits, and should provide guidance for disentangling complicated reaction norms generated in life history trade-off experiments.
back to schedule
Wednesday October 23rd at 12:30 pm
Search strategies and encounter rates of interacting species.
Andrew Hein
Encounters between individual organisms are an essential part of biology. Population ecologists since the time of Alfred Lotka have used kinetic equations to model the rates of such encounters. These equations form a basic component of coupled population dynamics. In this talk, I'll explore the connection between ecological encounter rate kinetics and organismal search behavior. I'll show that the mathematical tools ecologists typically use to model encounter rates do not include important features of organismal search behavior. I'll suggest several ways to include sensing and search behavior when deriving encounter rate models, and show that this can lead to new encounter rate kinetics.
back to schedule
Thursday October 24rd at 11:30 am
Agent-based modeling in Natural Resource Management
Emilie Lindkvist
This talk will focus on the two main research strands I'm currently working on with Jon Norberg and Maja Schlüter at the Stockholm Resilience Center. Both projects address natural resource management issues using a classic population growth model, but focus on different aspects of human-environment interactions.
1) Learning-by-doing (LBD): This study is a theoretical experiment, examining an isolated LBD process, where a manager of a fish population learns and makes decisions based on his perceived reward and stock assessment, with the aim of finding the maximum economic yield over time. We investigate the efficiency of LBD relative to resource complexity. Questions addressed are 1) what are adequate degrees of experimentation vs. exploitation and level of discounting? 2) How fast should the manger adapt his view of the system in relation to perceived feedback changes? 3) When is hindsight relevant? To operationalize this we created an artificial intelligent LBD agent equipped with state-of-the-art learning features that exploits a resource with different characteristics.
2) Collective Action. Here we study the evolution of cooperative versus non-cooperative strategies for self-governance in small-scale fisheries. We inform an agent-based model with empirical data from Kino Bay, Mexico, to investigate under which social and ecological conditions these two forms of self-governance emerge. We are particularly interested in the co-evolutionary dynamics that arise from the interactions between fishermen, fish-buyers, and the fish population they exploit. The presentation will focus on the conceptualization of the fishery dynamics and the key processes in the model.
back to schedule
Wednesday November 6th at 12:30 pm
Joint evolution of dispersal and local adaptation in heterogeneous environments
Andrew Berdahl
Both dispersal and local adaptation are major drivers of the structure of populations; however, in general, their respective roles are not independent and the two may often be at odds with one another evolutionarily, each one exhibiting negative feedback on the evolution of the other. In this study we investigate their joint evolution by letting both evolve simultaneously under natural selection in a simple meta-population model. We show that, depending on environmental characteristics, their interplay leads to either a monomorphic population of highly dispersing generalists or a rarely dispersing, locally adapted, polymorphic population. A critical value of environmental heterogeneity divides these two selection regimes, around which we observe discontinuities and hysteresis in the evolutionary stable strategies.
back to schedule
Monday November 11th at 11:00 am
Are social populations evolvable?
Alexander Stewart
Social interactions abound in nature, ranging from humans, to honey-bees, to bacteria. But what happens when a social population faces an environmental change, and must acquire an adaptive mutation -- does the frequency dependence of social types on each other facilitate or impede adaptation? One the one hand, social interactions can promote greater standing genetic diversity, which may accelerate the arrival of a beneficial mutation. But, on the other hand, if social interactions require the maintenance of multiple, cooperative types, then each such type will need to fix the beneficial mutation, so that adaptation might be impeded. Here we model the effects of mutually beneficial, cooperative social interactions on the evolvability of populations. We show that social interactions can either accelerate or retard adaptation to a novel environment, by an order of magnitude, compared to populations that lack social interactions. We quantify how the rate of adaptation depends on the number of different social interactions maintained in the population, and on the number of mutations required for adaptation. We find that that, in many regimes, populations maintaining multiple different cooperative social interactions are able to fix "complex" multi-step adaptations much more quickly than asocial populations.
back to schedule
Wednesday November 13th at 12:30 pm (to 2:00 pm)
On world trade, social physics and the other shore of complexity
Matthieu Barbier
Econophysics and similar recent brands of cross-disciplinarity have been quite variably received in their target communities, especially for some radical attempts at simplifying complexity. From working on models of international trade and language change, I have grown to believe and will try to demonstrate that even the barest of "toy models" could prove useful to the self-respecting economist or social scientist, if correctly integrated in a broader methodology. On a more speculative note, I feel that even the boldest of these endeavors have fallen short of pursuing their most interesting goal: the "reductive closure" of collective phenomena. Their inspirations in physics (notably equilibrium statistical mechanics and kinetic theories of fluids) not only allow to compute or simulate plausible behaviors, but underlie autonomous macroscopic theories, such as thermo- and hydrodynamics: closed sets of global variables and relations that may be used and extended without referring to the microstates. I will suggest how this process relates to multiscale modeling, and conclude with some practical ideas on how one might at least approximate it in socio-ecological systems, even if one, unfortunately, is not Ludwig Boltzmann.
back to schedule
Wednesday November 13th at 12:30 pm (to 2:00 pm)
Change in Nonlinear Dynamics and Spatial Structure of Coastal Socio-Ecological Systems: The Bay of Fundy as Case Study
Emily Klein
While there is growing consensus for ecosystem-based management of the oceans, it is critical to avoid isolating marine systems in time. As contemporary research demonstrates, the past is integral to comprehending long term change in these environments. Furthermore, if system health is a goal for the future, appreciation of conditions prior to intense anthropogenic pressures can offer unique insight into ecosystem structure, function, and dynamics. However, to date, this research has been limited to equilibrium and linear views. Other work is increasingly demonstrating the prevalence of nonlinear behavior in the oceans, and challenging conventional assumptions of linearity and equilibrium. My doctoral research is the first to explore nonlinearity in an ecologically and economically vibrant, less degraded ecosystem of the past. Investigations combined qualitative narratives, novel analytical approaches, and historical data (circa 1870-1920) to expand knowledge on ecosystem dynamics, spatial structure, and resilience for the Bay of Fundy. Findings suggest the previous Bay ecosystem was highly nonlinear and locally diverse, yet interconnected with local substocks and potential subsystems. This structure may have been critical for population persistence and system resilience. Finally, although human influence clearly affected species abundance and distribution, it also shifted from local to regional scales. These spatial impacts likely caused similar change in the ecosystem, fundamentally undermining system resilience that had previously supported the Bay during centuries of human use.
back to schedule
Wednesday November 20th at 12:30 pm
Profit spirals and extinction shadows: Predicting and preventing extinction threats from overharvesting
Matthew Burgess
Overharvesting poses a significant threat to biodiversity, particularly marine biodiversity. Threats to populations from overharvesting are commonly assessed phenomenologically, by estimating past population decline rates, current rarity or current harvest rates, which tends to identify already declining or collapsed populations, whose recoveries can be slow and costly. I illustrate how simple mechanistic theories can be used to empirically identify combinations of biological and socioeconomic conditions that could eventually cause a population to be driven extinct or severely depleted, before dramatic population declines and high harvest rates are realized. Mechanisms of extinction by overharvesting often fall into one of two categories: 'profit spirals' - where low costs and/or price increases prevent profits from declining as a harvested population is depleted - and 'extinction shadows' - where the ecological or economic characteristics of some harvested species indirectly facilitate the extinction of other species they are caught or sold with, or interact with ecologically. I present simple approaches to empirically predicting future threats of extinction and severe depletion from these mechanisms, and applications of these approaches to tunas and billfish. I discuss possible interventions to prevent fisheries and other harvests from causing extinctions and severe population declines.
back to schedule
Wednesday December 4th at 12:30 pm
Modeling Mate-finding by the Plain Ringlet Butterfly: ... Random Walks, Lévi Walks, and Worse
Henry Horn
I develop agent-based NetLogo models to explore varied random walks, and I fit those models to metrics of the Plain Ringlet's environment and behavior. Adding demography, I estimate lifetime encounters with potential mates as a function of a male's time spent in active "search."
The models are variations on: repeated one-step blind-man's-buff, Lévi walk, and forward-biased random walk. The environment may be: unbounded, toroidal, or bounded with varied edge-behavior. Contrasts among the models provide general insights into: strategy and tactics of searching, competition for first access to resources, effects of boundary geometry and behavior, tests of rules of movement, and even conditions under which habitat fragmentation may be a good thing.
This is work in progress. Some results are so new that I haven't had them yet.
back to schedule
Tuesday March 11th at 2:00 pm
Title not yet available.
Rick Durrett
Abstract not yet available.
back to schedule
[back to schedule]
Links to previous schedules
Fall 2000
Spring 2001
Fall 2001
Spring 2002
Fall 2002
Spring 2003
Fall 2003
Spring 2004
Fall 2004
Spring 2005
Fall 2005
Spring 2007
Fall 2007
Spring 2008
Fall 2008
Spring 2009
Fall 2009
Spring 2010
Fall 2010
Spring 2011
Fall 2011
Spring 2012
Fall 2012
Spring 2013
| 