Phase IV

1
Phase IV Bayesian Learning Reloaded

• Operator Eric Bengfort
• Temporal Status End of Week Eight
• Location Phase Four Presentation
• Systems Check Cleared
• Status Entity its alive!

2
Previously

• Used Bayesian Learning to probabilistically
  classify unseen objects based upon their features.

Testing Data
Training Data
Probabilistic Categorization
Lets do something interesting with this
-------------------gt
3
Idea Goals

• Use Bayesian Learning to govern how an entity
  behaves in an environment.
• Let the classification classes be actions the
  entity is able to perform.
• Give the entity minimal instinct to start with,
  and then set it loose.
• User is able to praise or scold entity to
  reinforce proper behavior.
• Write program generically so that an end user can
  create any entity with any potential actions
  inside any environment.
• Do this in a week.

4
Data Driven Everything

• Everyone loves puppies, lets make a puppy!
• Puppies have to learn how to behave in real life.
• Three text files will build our puppy and the
  world.

entity.txt Name of entity. instinct.txt
Entitys starting intelligence. Must have at
least one example of each possible
action/class. worldObjects.txt Objects entity
engages.
5
Profit Margins

• By merely editing text files, unlimited scenarios
  can be generated and experimented upon.
• Demonstrates the power of Bayesian Learning in
  real time. The response from the user
  (praise/scold) directly influences entity
  development.
• My personal testing greatly exceeded personal
  expectations.

6
Launch the demo already!!

• What I expected.
• More talk about the internals of the program.
• The surprising results.