Knowledge representation

Knowledge representation

1. Human brain - Long-term memory analogy

   • Declarative Memory
     • Associated with knowing "what things are" i.e., facts
   • Procedural Memory
     • Associated with knowing "how to do things" i.e., Methods

2. Implicit Knowledge

   • Neural networks
     • Used in powerful pattern matchers
     • Agent trains and learns in new situations
       • Sensitive to training set
       • Implicit formation of knowledge by activation
   • Behavioural based
     • Rodney Brooks
       • Post-experience planning
     • Emergent behaviour
       • Simple agents produce a non-explicitly programmed complex behaviour
       • Analogies
         • Birds swarming
         • Ants co-operating
   • Chunking
     • Problem decomposition
       • Solve & cache solutions to sub-problems
         • Combine cached solutions forming one solution

3. Explicit Knowledge

   • GOFAI
     • Knowledge of World
       • Observed, acquired, interrogated & revealed
       • Engineered, reasoned, used & deducted
       • Hierarchical planning
         • Meta-level: scheduling, planning, ordering
     • Lacks & constraints
       • Reactive planning & problem solving
       • Environmental needs
         • Maximum observability
           • All aspects relevant to choice of action observable
         • Deterministic environment
           • Saves worrying about uncertainty
         • Static environment
           • Environment doesn't change whilst agent delibrates
       • Adaptability
   • Facts & Rules
   • Data Structures
     • Semantic nets
       • Symbolic representation of relationships between concepts
     • Frames
       • Hierarchical structure
         • Classes, sub-classes
         • Instances (e.g., name)
         • Slot:fillers(i.e., description of attributes)