Representations of single and compound stimuli in negative and positive patterning

Abstract

In four experiments, rats were trained on different patterning discriminations before being tested with comp pounds composed of novel combinations of the trained stimuli. In Experiment 1, rats were trained on a negativep patterning schedule (A B AB ) intermixed with reinforced presentations of a second compound (CD ). On a subsequent test, the rats responded more to two novel compounds, AC and BD, than to A and B, but less than to CD. In Experiment 2, rats were trained on two concurrent negative-patterning discriminations (A B AB , C D CD ). On test, they responded more to AC and BD than to AB and CD, but less than to the single stimuli. In Experiment 3, rats were trained on two concurrent positive-patterning discriminations (A B AB , C D CD ). On test, their response rates to AC and BD were not different from the response rates to the trained compounds (AB and CD). Finally, in Experiment 4, rats were trained on a positive-and negativepatterning discrimination concurrently. Once again, on test, response rates to AC and BD were not different from patterning discrimination concurrently. Once again, on test, response rates to AC and BD were not different from responding on reinforced trials of the trained discriminations (A , B , and CD ). We discuss the implications of these findings for elemental and configural models of stimulus representation. Learning & Behavior 2009, 37 (3), [230][231][232][233][234][235][236][237][238][239][240][241][242][243][244][245] J. A. Harris, justinh@psych.usyd.edu.au S STIMULU TIMULUS S REPRES SENTATION ENTATIONS S IN IN PATTERNING NING R R DIS SC CRIMINATION IMINATIONS S 231 231(CD ). After extended training (60 days), they were tested with these four single and compound stimuli, as well as with two new compounds (AC and BD) and the single stimuli (C and D; none of the new stimuli were reinforced during the test). This design is similar to that described by Rescorla (1972) and Whitlow and Wagner (1972), the key difference being that the rats were trained with a compound (CD ), rather than with a single, CS (C ). The reason for this change was that the design of the earlier experiments (A B AB C ) creates a correlation between stimulus number (single vs. compound) and reinforcement, and if the rats learned this correlation (i.e., that compound trials are not reinforced), it would bias responding to the novel compounds on test. This might occur if the rats could use number or any property correlated with number (e.g., overall stimulation intensity) as a configural element that would readily generalize from AB to the test compounds. By training CD rather than C , the present experiment removed this correlation, because AB trials during training were balanced by CD trials. The Rescorla-Wagner model predicts that responding to AC (or BD) will be given by the summed associative strengths of A and C (or B and D). Because the associative strengths of A and B should be greater than those of C and D, the AC and BD compounds should have greater associative strength and, thus, elicit more vig...

An attention-modulated associative network

“…The model we propose here does a better job than its predecessors in accounting for the data presented by Harris et al (2009). As is shown in the right-hand panels in Figure 11, when the model is trained on the discriminations to the point at which the simulated activation of US elements matches the level of discrimination shown by the rats' behavioral responses, the model's prediction for the relative strength of responding to the novel compounds corresponds very closely to that observed empirically.…”

“…The stronger activation of those elements means that their associative strength is expressed in the new compound and, thus, responding is higher to AC than to AB. Indeed, given these processes, the model tends to predict greater responding to the new compounds than was actually observed in the experiments by Harris et al (2009). However, the model's prediction for response rate is reduced if it assumes that stimuli from the same modality contribute to one another's normalization, as we have assumed in the simulations presented in Figure 11.…”

“…After extended training, the rats were tested for responding to the novel compounds AC and BD. The rats responded to these compounds significantly more than to AB and significantly less than to CD; their responses to AC and BD were above, but not significantly different from, their responses to A and B (Harris et al, 2009). In a second experiment, rats were trained on two concurrent negativepatterning discriminations (A B vs. AB , and C D vs. CD ), before being tested with the novel compounds AC and BD.…”

“…The test data from all three experiments are presented in Figure 11. As was discussed by Harris et al (2009), each of the earlier models discussed here can account for the data from each of these experiments individually, but only by adopting different parameters in each case. That is, none of the models previously considered was able to simultaneously explain the data from all three experiments if constrained to adopt the same set of parameters across experiments-a reasonable requirement, given that each experiment used the same set of for recent reviews of latent inhibition, see Holmes & Harris, in press;Lubow & Weiner, 2010).…”

Olfactory learning without the mushroom bodies: Spiking neural network models of the honeybee lateral antennal lobe tract reveal its capacities in odour memory tasks of varied complexities

“…However, single-layer perceptrons cannot solve the XOR problem because there are no planes that can be drawn across the space of inputs to separate the single components from their mixture. Numerous experimental studies have revealed better discrimination performance for positive patterning than for negative patterning [74–76]. However, a feed-forward network containing hidden units (multiple layers) can classify any inputs [77–78].…”