Next: Backprop in Action
Up: Neural Network Learning
Previous: Hypothesis Space & Inductive
- Backprop can discover useful intermediate representations at the
hidden unit layers
- way to make implicit concepts explicit (not across rows!!!)
- discovering binary encoding
- important degree of flexibility
- more layers of units - more complex features can be invented
Patricia Riddle
Fri May 15 13:00:36 NZST 1998