1 Self-Organization Self-organization is a process in which - PDF document

Umeå University Department of Computing Science Emergent systems Spring-12 Self-organization, autonomous agents and ant algorithms http://www.cs.umu.se/kurser/5DV017 Previous lecture ❒ Nonlinear dynamic systems ❍ The Logistic map ❒ Strange attractors ❍ The Hénon attractor ❍ The Lorenz attractor ❒ Producer-consumer dynamics ❍ Equation-based modeling ❍ Individual-based modeling 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Outline for this lecture ❒ Self-Organization ❒ Autonomous Agents ❒ Real Ants ❒ Virtual Ants ❒ Ant Algorithms 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 1

Self-Organization ❒ ”Self-organization is a process in which pattern at the global level of a system emerges solely from numerous interactions among the lower-level components of the system. Moreover, the rules specifying interactions among the system’s components are executed using only local information, without reference to the global pattern.” – Camazine et al, p. 8 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Self-Organization ❒ Pattern ❍ A particular, organized arrangement of objects in space or time ❒ Interactions ❍ Based on local information only - no global information ❍ Physical laws ❍ Genetically controlled properties of the components 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Self-Organization - Ingredients ❒ Positive feedback ❍ Activity amplification ❒ Negative feedback ❍ Activity balancing ❒ Amplification of random fluctuations ❒ Multiple interactions 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 2

Self-Organization - Information ❒ Signals ❍ Stimuli shaped by natural selection specifically to convey information ❒ Cues ❍ Stimuli that convey information only incidentally ❒ Gathered from one’s neighbors ❍ Stimuli-response, simple behavioral rules of thumb ❒ Gathered from work in progress ❍ Stigmergy ❍ Random fluctuation and chance heterogeneities 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Self-Organization - Characteristics ❒ Dynamic systems ❒ Exhibit emergent properties ❍ Attractors ❍ Multistability ❍ Bifurcations ❍ Parameter tuning ❍ Environmental factors ❒ Adaptive systems ❒ Different patterns may result from the same mechanism ❒ Simple rules, complex patterns 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Self-Organization – Alternatives ❒ Central leader ❍ Need effective communication and cognitive abilities ❒ Blueprints ❍ Must be stored ❒ Recipes ❍ Hinders flexibility ❒ Templates ❍ Must be available 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 3

Stigmergy ❒ A recursive control system ❒ Effective for coordination in space and time ❒ A sequence of qualitatively different stimulus-response behaviors ❒ Two types: ❍ Qualitative stigmergy ❍ Quantitative stigmergy 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Stigmergy - Advantages ❒ Permits simpler agents ❒ Decrease direct communication between agents ❒ Incremental improvement ❒ Flexible, since when environment changes, agents respond appropriately 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Autonomous Agent ❒ ”a unit that interacts with its environment (which probably consists of other agents) ❒ but acts independently from all other agents in that it does not take commands from some seen or unseen leader, ❒ nor does an agent have some idea of a global plan that it should be following.” - Flake, p. 261 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 4

Real Ants ❒ Imagine if artificial systems could do the things ants can do? ❒ Why ants? ❍ Amazonas: 30% of biomass is ants/termites ❍ Amazonas: dry weight of social insects is four times that of other land animals ❍ Earth: ~10% of total biomass (like humans) 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Army Ants ❒ 100 000s in colony ❒ Create temporary ”bivouacs” ❒ Act like unified entity (Pictures from AntColony.org) 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Fungus-Growing Ants ❒ "A Leaf Cutter Colony can strip the tallest of trees in a single day. Equivalent consumption of a full grown cow in the same time!" ❒ ”Cultivate” fungi underground ❒ Fertilize with compost from chewed leaves (Pictures from AntColony.org) 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 5

Fungus Cultivator Nest (Picture from AntColony.org) 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Langton’s Virtual Ants ❒ Grid with white or black squares ❒ Virtual ants can face N, S, E, W ❒ Behavioral rule: ❍ Take a step forward ❍ if on a white square then paint it black and turn 90º right ❍ if on a black square then paint it white and turn 90º left 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Virtual Ants - Example 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 6

Virtual Ants – Time Reversibility ❒ Virtual ants are time-reversible ❒ But, time-reversibility does not imply global simplicity ❒ Even a single virtual ant interacts with its own prior history ❒ Demonstration 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Virtual Ants - Conclusion ❒ Even simple, reversible local behavior can lead to complex global behavior ❒ Such complex behavior may create structures as well as apparently random behavior 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Harvester Ants ❒ Find shortest path to food ❒ Prioritize food sources based on distance and ease of access (Picture from The Texas A&M University System) 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 7

Adaptive Path Optimization ❒ How do they do it? ❍ Deposit pheromone • Can be several different • Can detect gradients and frequency of contact ❍ Does not follow trails perfectly • Exploration ❍ Feedback reinforces ”good” trails 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Adaptive Path Optimization ❒ Adaptive significance ❍ Chooses the “best” food source ❍ Chooses the shortest trail ❍ Adapt grade of exploration to the quality of the food source ❍ Collective decision making ❒ Observations at trail formation ❍ If equal length, one is chosen randomly ❍ Sometimes a longer/worse is selected ❍ Pros • Easier to follow • Easier to protect • Safer 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Formation of trails ❒ Find trail ❍ ”Forager” deposit pheromone ❍ How and when pheromone is deposited varies ❍ Other follows trail ❍ Pheromone also act as orientation aid ❒ Follow trail ❍ P L = (( C L + k ) h ) /(( C L + k ) h + ( C R + k ) h ) ❍ C L , C R : concentration of pheromone ❍ k , h : to fit the model to experimental data ❒ Pheromone evaporation ❍ Trails can last several hours to several months ❍ The lifetime of pheromone • average 30-60 min, but can be detected much longer 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 8

Ant Algorithms ❒ Basic ingredients for all ant based algorithms ❍ Positive feedback • Reinforce good solutions • Reinforce good parts of solutions • Through pheromone accumulation ❍ Negative feedback • Avoid too early convergence • Through pheromone evaporation ❍ Cooperation • Parallel search • Through more ants and through pheromone trails 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Ant Algorithms ❒ Ant colony optimization (ACO) ❒ Developed in 1991 by Dorigo (PhD dissertation) in collaboration with Colorni and Maniezzo 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU Summary ❒ Self-Organization ❒ Autonomous Agents ❒ Real Ants ❒ Virtual Ants ❒ Ant Algorithms 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 9

Next time ❒ Flocks, Herds, and Schools ❒ Boids 28/1 - 13 Emergent Systems, Jonny Pettersson, UmU 10