Lang and I softened (made more robust) the background model in our brute-force foreground–background modeling system; that is, we made it harder for the foreground model to pick up spurious streams by effectively modeling deficiencies in the background model. This worked like a charm, eliminating almost all streams in our "scramble" sample and leaving a few tantalizing ones in the real data. Now we have to figure out how to observationally confirm our candidates. This is hard, because when you have used all the data to find the candidates, you don't have any data left to confirm them!
No comments:
Post a Comment