-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Artificial Intelligence
- search (overlaps with algorithms)
  breadth first search
  depth first search
  minimax
- logic (overlaps with logic, but not allowed to use same methods)
  first order logic
  unification 276 (read deductive foundations of comp sci, not russel/norvig)
  skolemization 296
  resolution
  1.) elimintate implications: p => q becomes -p v q
  2.) move - inwards: -Axp becomes Ex-p and -Exp becomes Ax-p
  3.) standardize variables
  4.) skolemize: remove Ey in AxEy ... replace y with f(x) (or f() or f(x,x'))
  5.) drop universtal quanitifers
  6.) distribute ^ over v (AND over OR)
  now in CNF
  resolution refutation
    assume negation of what you want to prove, then derived contradiction
  resolution rule example
    l1 or l2, -l2 or l3
      therefore
        l1 or l3
    l2 and -l2 are complementary literals
  resolution rule is enough for complete inference algorithm
- probability
  Bayes - p(ab) = p(b)p(a|b) = p(a)p(b|a)
- review summation mathemetics
- belief networks
- natural language (grammars)
  (overlap with programming languages, compilers, automata)
- perceptrons
  weighted inputs, threshhold on output
- learning
  decision trees
  naive bayes classifier
- language
  lexicon of Eo: classified words
  grammar of Eo: rules for combining lexicon (context free, see augmented)
  parsing
  derivation
  augmented grammar 806
    adds context to context free grammar
    definite clause grammar
      convert
        X -> Y Z 
        Y(s1) ^ Z(s2) => X(s1+s2)
  semantic interpretation 813
- vision (segmentation)