Artificial neural networks as models of stimulus control*

Figure 4: Generalization gradients for rats responding to different click rates. Continuous
line: control group, no discrimination training; dashed line: S⁺ at 32 Hz and S^- at 20 Hz;
dotted line: S⁺ at 20 Hz and S^- at 32 Hz. Data from Weiss & Schindler (1981, experiment
one). On the vertical axis relative percentages of responding are plotted.

were reinforced for responding to click rates of 32 Hz, and not reinforced at a
lower click rate, 20 Hz. The resulting gradient (line 2 in figure 4) is similar to
those in figure 3, with a strong peak-shift in the direction of increasing click rate
and responding dropping well below that of the control group for stimuli less in-
tense than S⁺ .

The most interesting feature of this study is the reversed gradient (line 3 in
figure 4) obtained from rats trained to respond to the lower click rate of 20 Hz
as opposed to the nonrewarded, and more intense, 32-Hz stimulus. From these
data we see clearly that it is not intensity in itself that is responsible for increased
response rates, but rather the interplay of all experiences that animals are met with
along the considered intensity continuum.

All these findings are well reproduced by the network model, as seen in fig-
ure 5. The networks exhibit also a consistent negative peak-shift: stimuli that are
further than s^- from s⁺ elicit less responding than s^- itself, not being important
which one between s⁺ and s^- is the most intense. Does this occur with animals?
In all of the cases in figure 3 the S⁺ was rewarded in contrast with the absence of
stimulation, i.e. no light, bell or whistle, so that investigation with stimuli less in-
tense than S^- could not be conducted. A different kind of floor effect, i.e. very low
responding to S^-, prevents us from testing the existence of a negative peak-shift
for data in figure 4. The existence of such a phenomenon is, however, well estab-
lished in other experiments (e.g. Thomas & Setzer, 1972; Zielinski & Jakubowska,
1977), although this effect has not been the object of systematic investigations.

<<   5   6   7   8   9   10   11   >>

More intriguing information