Finding It Now: Construction and Configuration of Networked Classifiers in Real-Time Stream Mining Systems - Signal Processing Systems - page 122

Fig. 13 Time scales for decentralized algorithms

Decentralized Algorithm 3 Partial Search Ordering Algorithm— for classifier C i =
C σ ( h )

1. Observe state )

2. With probability p i j , request utility parameters v σ ( h + 1 )

( θ i , Children ( C i )

= v j w j for any of w σ ( h + 1 )

)
3. For each child probed, compute corresponding utility
4. U i (

the N − h classifiers C j ∈ Children ( C i

+ v j w j T i t h − 1
g h − 1

0

C j )= − ρ σ ( i )

5. Select the child classifier with the highest U i as trusted child .

6. Compute the corresponding v i w i and transmit it to a previous classifier who requested
it.

4.3.3 Decentralized Ordering and Operating Point Selection

In case of unfixed operating points, the local utility of classifier C i = C σ ( h ) also

depends on its local operating point x i —but it does not directly depend on the
operating points of other classifiers 7 :

x i ) t h − 1
g h − 1

U i = − ρ i 0 + v h + 1 w h + 1 T i ( .

As a consequence, we can easily adapt the Partial Search Ordering Algorithm
into a Partial Search Ordering and Operating Point Selection Algorithm by comput-
ing the maximal utility (in terms of x i ) for each child:

7 The utility parameters v j w j fed back from classifier C j to classifier C i are independent of any
classifiers' operating points.