12 06 06 evening discussion

General level:

Look at how the structure of networks can influence the evolution of certain types of

behavior such as cooperative behavior or information sharing and ?

Flip side, how do rules for generating network structure evolve, and can you get a

coevolution between the two processes.

Individuals can evolve by the fitness they recieve, but one can think of the outcomes

for the group as a whole, eg, does it become more efficient, do autonomous network

structures emerge. Super connected components, subgroups, interelation.

Applicability of these concepts to invididual interests?

Baboons: Tradeoff between protection from predation in large groups versus less

competition for resources in small groups.

Connection between social and spatial relationships. Add this later, if we have time.

Too complex to have both dimensions to begin with.

Issues for each topic: Idea Analogies Literature Tools

How does network structure influence cooperation, how does individual behavior

evolve in this context?

Two steps: 1. How does a fixed network influence individual behavior? 2. How do the behaviors that create that network evolve 1+2. How do you make these influences fit together?

Network of neurons which alter their connections as well as being altered by signals

across the network.

Speed of evolution/adaptation. Timescales on which behavior and network structure

change.

Metric for measuring individual behavior versus network structure.

Cartoon version: Agents are nodes Modes are playing games with one another They can form or break ties

What are we looking for? - Steady state? - Dynamics?

At what level to measure performance? - nodes only? Individual fitness function individuals are trying to maxmise - network level as well - different metrics? Group level fitness function

Environment: - foraging - heterogenous environment? - pure game - endowment - cf uniform resource landscape

Two simple start points: - heterogenous landscape, foraging problem, attention to social cues - uniform landscape, more complex interactions - games

Can we create a general model where pushing different parameters to zero gives

these different models?

Timescales - generations - natural selection Non-overlapping generations and different networks forming each generation, or Overlapping generations and individuals coming into an existing network.

If there's a *cost* to signalling some information - cos others will compete for food,

or there's a time cost to signalling

What benefit could there be to signalling the presence of a resource? Reciprocity,

population level success. The population level outcome may be better with signalling

but it may not be an ESS.

1. All individuals have the same rules for making and breaking connections - measure

metric of whole group success - group selection with indentical individuals within

groups. 2. Evolve individual strategies

Stratgies: 1. to form connections 2. Whether to share or not across the networks you have

Existing lit: Games on nets - eg UG with random cutting and forming of new links

Learning - an effect of an individual's history in the game on how they behave in the

game

It would be interesting to compare evolutionary learning with within lifetime

reinforcement learning.

How to allow individual learning rules to evolve?

Possible further complexification. Structure of game - Agents meet and decide on a

game from an ensemble of possible games, then play according to their strategies for

those games

Alternatively run the same model with different games in different runs.

Which games? PD PGG - is there previous work on PGG on networks?

Group size issue?

How to split the work up? Develop six abstracts (one each) see whether different ppl

want to do different parts/aspects?

Foraging model - space as a weighted network of interaction probabilities.

Reorganisation of network means movement in space. Game is a foraging game this is

a specific instance of the general structure we've specified.

Tools:

There are utilities from UMich to automate experiments and so on.

Java, RePast - Matina, Jonathan, Me, Juan C++ - Greg

Literature:

What have ppl done already

Play a PD with your neighbours, kill an edge with someone if they defect on you, and

form a new one at random. Reproduce after a number of rounds according to total

payoff.

Jonathan has a lead on some analytic work on games on different structures.

Is there previous work on PGGs on nets?

TODO:

Lit search + three lines for each paper