|
|
|
|
|
|
|
|
|
• |
Training
using updating sequence in
|
|
|
reverse
order speeds convergence.
|
|
|
• |
Tradeoff:
Requires more memory
|
|
|
• |
Suppose
exploration and learning cost great
|
|
time/expense.
|
|
|
• |
Can
retrain on same data repeatedly.
|
|
|
• |
Ratio
of old/new update sequences a matter
|
|
|
of
relative costs for problem domain.
|
|