next up previous
Next: 5.1 Possible Concerns with Up: Accelerating Reinforcement Learning by Previous: 4.4 Robot ArmNew

5 Analysis of Results

 

The experiments of the previous section have shown that function composition produces a significant speed up across two different types of related task and across two domains. In addition, the composed solutions tend to be very accurate and little further refinement is required. This section begins by looking at possible concerns with the experimental methodology that might affect the measurement of speed up. It then discusses various properties of the task being solved that affect the speed up achieved by using function composition.





Chris Drummond
Thursday January 31 01:30:31 EST 2002