next up previous contents
Next: Interference and old data Up: Pros and cons vs. Previous: Training and validation

Prediction speed

Neural networks make up for some of their slowness in training with fast prediction speed. As mentioned above, the prediction speed of memory based learners can suffer as the dimensionality increases. One solution to this problem is the efficient tree structure used by Vizier. It allows for fast finding of nearest neighbors and grouping of nearby data points for faster weighting and retrieval of information. These methods are described later in the section on kd-trees.



Jeff Schneider
Fri Feb 7 18:00:08 EST 1997