Conclusion

• MLPs work. Use them for machine learning problems of moderate complexity.
• Take Home Points
  1. Perceptrons have inputs that are typically weighted.
  2. There are a range of transfer functions, including linear, sigmoid, and step.
  3. One layer of perceptrons can't learn some simple things by themselves, so we use MLPs.
  4. MLPs can represent almost any function.
  5. Using backpropagation, they can learn any function.
• Reading: For this week is the MLP Wiki.
• Reading: for next week is Russell and Norvig 18.1 through 18.3 (pp 704 - 718 Learning from Examples)