How to `program' neural networks?

• The learning problem

• Selfridge (1958): evolutionary or `shake-and-check' (hill climbing)

• Two main approaches in connectionism

  • Hebbian learning
  • Error-correction learning