How to `program' neural networks?

• The learning problem

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

• Other approaches

  • Unsupervised or regularity detection
  • Supervised learning
  • Reinforcement learning has `some' supervision