2 |
Few Tracks with Dirty Data
-
Now we want to view total tour (track) as a single entity and use either annealing or elastic net methods?
-
Kalman filter fails as it uses information incrementally - adds one time slice to previous local track
-
c2 gets stuck in a local minimum - need concept of temperature to avoid false minima
-
Neural networks inefficient as too much redundancy
-
Deterministic Annealing (elastic net) is a good possibility
-
Rose has tested this
-
Note in tracking, tracks are attracted to measurements;
-
in navigation, vehicles are repelled from obstacles, otherwise identical
|