Page 299 - Deep Learning

282 Adaptation

(a) 35
Single Simulated Subject
30
2
R =1
25
Errors per Trial 20

15

10

5

0
0 2 4 6 8 10
Trial
(b) 2000
Sum of 50 Simulated Subjects
2
R =0.99
1500
Errors per Trial 1000

500

0
0 5 10 15 20 25 30
Trial
Figure 8.6. The effect of additive aggregation on the shape of the error reduction
curve. Panel (a) shows a simulated subject with perfect fit to an exponential curve.
Panel (b) shows the curve for the sum of the performance of 50 simulated subjects.

curve of the shape we would expect from the prior analysis. Other transportation
industries analyzed by Duffey and co-workers conform to the same pattern.
The additive model is applicable to the case of populations, defined here
as sets of nearly independent operators. However, this case is oversimplified.
It is more common that the agents in a social system interact in ways that