Serendip is an independent site partnering with faculty at multiple colleges and universities around the world. Happy exploring!

Reply to comment

Emergence 2009: Final Projects, II

Emergence

Biology 361 = Computer Science 361
Bryn Mawr College, Spring 2009

 

Back to index of models

 

 

Download/view: LearningModel3.nlogo

 

WHAT IS IT?

This aim of this model is to show different types of learning:

1) learning by experience and

2) learning through direct interaction (cultural learning).

The general idea is that agents that learn by experience engage in behavior that consists of making less mistakes. Learning by interaction results in a type of learning behavior that requires time in order to get the desired avoidance behavior. This is a relatively simple model that highlights the fact that not all learning behavior is the same and that in some types of environments, a type of learning has an advantage over another that can diminish over time.

HOW IT WORKS

What agents need to do:

-move, with some degree of randomness:
*the turtles all begin by moving around and navigating randomly
-observe, interact with world:
*this is present in my model in the turtles, which have to observe the patch color that they are on as they continue to navigate and proceed onto interacting with the world (the rest of the turtles)
-learn from interactions with world:
*this is the main point or objective of my model learning by experience and learning by interaction (cultural learning)
-get it less wrong - induction
*learn to associate red barriers as something that should be avoided adopt avoidance behavior so that eventually both types of learners are essentially acting and navigating in the same way in this environment.The more interactions you have and the more you are exposed to something, the more likely you are to get it wrong the next time.
**again, this is a very simple model and may not apply to all types of learning (such as fear and other things that become reinforced very quickly)
-create internal models of the world
*one in which the turtles learn to avoid barriers, even the ones that initially learned by experience

HOW TO USE IT

-setup button set all the starting conditions

-hit go to let the model start running

-hit the number-turtles button to set the number of turtles

THINGS TO NOTICE

Notice:

1) the starting color of the turtles

2) the different colored turtles that emerge, two types: Blue turtles and yellow turtles.

3) Notice the avoidance behavior of the turtles

4) The mistakes rate graph: look at how over time the number of mistakes for both groups of turtles change. What happens? The mistakes of the blue turtles decrease and become similar to those of the yellow turtles (direct experience learners)

THINGS TO TRY

You can try:

-adding varying the number of turtles to see how long it takes for the interaction learners and the direct experience learners to perform in the same way.
-add less turtles to closely watch the behavior of the turtles and the time when they change colors
-the TRACE button allows for one to see the paths that the turtles have taken
-the ESCAPE graph: keeps track of the number of BLUE (interaction learners) turtles that have left the central rectangle area. This graph only shows blue escapees, NOT yellow. Future models can include a graph for yellow escapees since they also escape and are not perfect in their avoidance behavior.

EXTENDING THE MODEL
Several suggestions for future things to try:
-adding a button that changes the message value per interaction. If you count each interaction as 1 message, the model quickly reaches a point in which the blue turtles have interacted more than 10 times. Therefore, changing the number to a value less than 10 prolongs the probability behavior for the observer.
-alternatively, a reward could be given to turtles that engage in certain types of avoidance behaviors (again, there are many possibilites oh how to increase the complexity of avoidance behavior and learning over time)
-Could add a turtle that punishes other turtles to see how that affects learning behavior -Could create turtles with more traits to see how the interaction changes over time. Could also model different types of learning (ex: learning by fear of punishment/ health reduction)
-Could also change the color of the barriers to multiple colors so that the turtles are not only learning to avoid one color (red), but many.
-in order to allow for turtles to more effectively avoid red barriers, the random heading changes could instead be set to 90 or 180 degrees so that the turtles do not touch the red barriers at all.
CREDITS AND REFERENCES
Thanks to Professor Grobstein and Evan Raskin for their guidance and for helping me with the code for this model.

 

Models created using NetLogo.

Reply

The content of this field is kept private and will not be shown publicly.
To prevent automated spam submissions leave this field empty.
5 + 2 =
Solve this simple math problem and enter the result. E.g. for 1+3, enter 4.