| marp | true |
|---|---|
| theme | gaia |
| paginate | true |
| style | |
| math | mathjax |
--
- What is an agent (エージェント) about?
- An agent makes decisions or behaves following rules.
- An agent affects other agents or environments.
- Agent-based model
- An agent-based model simulates beased on behavior of agent(s)
- An agent-based model enables you to analyze holistic (全体的な) behavior of society by simulating the behavior, movement, action, and status of agents on computer.
- Example: Traffic simulation, Evacuation (避難) simulation in an emergency, spread on socials (SNS), collective behavior of organisms, artificial lives.
- There are mainly two types of agent.
- Movable agents move in environment.
- Immovable agents just make decisions without moving.
- Detailed models for agents do NOT always generate holistic behavior of society well.
- To build agent-based models, it is important how you abstract the objects.
- What is Game of Life (ライフゲーム) about?
- Game of Life models birth and death of lives.
- The black cells are alive, and the white ones are dead.
- The agents appear to move from cell to cell.
- You can check the regular creatures for Game of Life. -> https://conwaylife.com/wiki/
- Four basic rules for Game of Life
- Birth: A white cell turns black when it touches three black cells.
- Survival: A black cell stay black when it touches two or three black cells.
- Depopulation (過疎): A black cell turns white when it touches one or fewer black cell.
- Overpopulation (過密): A black cell turns white when it touches four or more black cells.
- Game of Life is a fundamental model with agents moving in discrete environment.
- Game of Life itself has few practical applications.
- However, the simple rules and initial conditions of Game of Life generate many vrieties of complex behavior, figures, and patterns.
- Game of Life provides a good example of modeling complex behavior with simple essential rules.
- Rule 30 of Game of Life generates a chaotic pattern, which is similar to the pattern of cone snails (イモガイ).
- Game of Life also suggests that complex systems in nature may be generated by simple rules.
- What is Boids about?
- Boids model flocks (群れ) of birds with a few simple rules.
- The agents move on a continous plane, not discrete.
- Three basic rules for Boids
- Separation: If any other agent in an area, an agent gets away from them.
- Alignment: If any other agent in an area, an agent moves in the same direction as them.
- Cohesion: If any other agent in an area, an agent gets close to their centroid (重心).
- Note that each area for the separation, alignment, and cohesion are defined separately.
- Boids are a fundamental model with agents moving in continous environment.
- Boids reproduce flocks of birds well with a few simple rules.
- Boids are applied to simulations and computer graphics (CG) for flocks of birds.
- Boids suggests that complex behavior of flocks, herds, and schools in nature are generated by simple rules.
- Many additional rules have been proposed for extended Boids, with obstacles, distinations, and randomization.
- Boids are applied to simulations for flow of people, schools of fish, drone shows, autonomous cars, and even artworks.
- What is prisoner's dilemma (囚人のジレンマ) about?
- Prisoner's dilemma models two prisoners cooperating and betraying.
- The agents just make decisions without moving.
- The agents gain benefits depending on the combination of their desion-making.
- Prisoner's dilemma is a fundamental model with agents making decisions without moving.
- Prisoner's dilemma models well the mismatches between socially optimal strategy and individually optimal strategy.
- Pareto efficiency (パレート最適) is a situation where one or more agents need(s) to decrease the profits to increase the profits of other agent(s).
- Nash equilibrium (ナッシュ均衡) is a situation where no agent could increase their profits by changing their own strategy.
- Prisoner's dilemma is applied to cooperative behaviors for animals and arms races (軍拡競争).
- You can check the regular strategies for iterated (繰り返し) prisoner's dilemma. -> https://plato.stanford.edu/entries/prisoner-dilemma/strategy-table.html
- The decision-making rules in this lecture are very simple.
- You can build a model, with multiple agents making decision by reinforcement learning.
- An exapmle of this combination is a hunting game.
- Predators and a prey move on a closed plane.
- The predetors need to hunt the prey together.
- The prey runs away from the predetors.
- The predetors learn strategies to hunt the prey by reinforcement learning, such as pincer movement (挟み撃ち) and ambushment (待ち伏せ).
