Artificial neural networks as models of stimulus control*



Figure 6: Gradients obtained in a light-wavelength generalization test with pigeons, repro-
duced from Hanson (1959). The lower gradient is a control one, obtained after training
on a 560-nm wavelength. The other gradients result from discrimination training between
the 560-nm light and other colours, as indicated in the legend.


Stimulus location

Figure 7: Network generalization gradients coming from the translation test. The s+ was
fixed, and four different negative stimuli were used, consisitng in a ‘signal’ like that in
s+ (see figure 2), but with a displaced position (indicated in the legend above). The evo-
lutionary algorithm gave entirely similar results, but for clarity only data from networks
trained with the back-propagation algorithm are shown. The input layer of the networks
was made of fifty units, and the positive stimulus was centered at position 30 (with a tar-
get output of 0.7) for the no-
s- group and all discrimination groups. The negative stimuli
(target output 0.2) were centered at the positions indicated in the legend.


10




More intriguing information

1. Outsourcing, Complementary Innovations and Growth
2. Konjunkturprognostiker unter Panik: Kommentar
3. ‘Goodwill is not enough’
4. The name is absent
5. The name is absent
6. Une nouvelle vision de l'économie (The knowledge society: a new approach of the economy)
7. Internationalization of Universities as Internationalization of Bildung
8. A MARKOVIAN APPROXIMATED SOLUTION TO A PORTFOLIO MANAGEMENT PROBLEM
9. Input-Output Analysis, Linear Programming and Modified Multipliers
10. Crime as a Social Cost of Poverty and Inequality: A Review Focusing on Developing Countries