Connectionist models of conditioning: A tutorial

Models containing networks of neuron-like units have become increasingly prominent in the study of both cognitive psychology and artificial intelligence. This article describes the basic features of connectionist models and provides an illustrative application to compound-stimulus effects in respondent conditioning. Connectionist models designed specifically for operant conditioning are not yet widely available, but some current learning algorithms for machine learning indicate that such models are feasible. Conversely, designers for machine learning appear to have recognized the value of behavioral principles in producing adaptive behavior in their creations.     

Appetitive differential conditioning of the rabbit’s jaw movement response

Two experiments examined appetitive differential conditioning of the rabbit’s jaw movement response (JMR) in a two-phase procedure. The first phase entailed reinforced training with one conditioned stimulus (CS+), and the second phase involved intermixed presentations of CS+ and an unreinforced stimulus (CS?). In Experiment 1, CS+ was a 600-Hz tone, and CS-was either a 660-,1,000-, or 2,100-Hz tone. In Experiment 2, the magnitude of the water unconditioned stimulus (US) paired with CS+ was either 1, 3, or 9 ml. The experiments revealed that 1) the level of responding to CS-rose for several days and then declined over the remainder of training; 2) the physical similarity between CS+ and CS?directly affected the level of responding to CS?but had no discernable effect on the level of responding to CS+ ; and 3) US magnitude positively affected the level of responding to CS + and, to a lesser extent, CS?. The results are discussed in terms of their implications for Spence’s gradient interaction theory and Pavlov’s external inhibition hypothesis.     

Blocking acquisition of the rabbit's nictitating membrane response to serial conditioned stimuli

Three experiments demonstrated that prior training with one stimulus (CS1) would block acquisition of the rabbit's nictitating membrane response (NMR) to another stimulus (CS2) which was more contiguous to the US during serial compound training (CS1-CS2-US). Specifically, the CS1-US interval was 800 msec, which produces only a modest rate of CR acquisition, while the CS2-US interval was 400 msec, which is an optimal value for the NMR preparation. Experiment 1 demonstrated blocking when CS1 overlapped CS2, and Experiment 3 demonstrated blocking when CS1 and CS2 were presented in a strictly sequential fashion. Experiment 2 showed that the magnitude of blocking in the serial compound was comparable to that obtained in a simultaneous compound in which both the CS1-US and CS2-US intervals were 800 msec, thus making CS2 less contiguous with the US than in the serial compound. Moreover, the level of responding to CS2 in all serial compound groups (blocking and control) was lower than in the simultaneous compound groups. The present findings provide further evidence that the associative consequences of CS-US contiguity can be highly attenuated by processes of attention or competition for associative strength.     

Temporal specificity in cross-modal transfer of the rabbit nictitating membrane response

Two experiments examine cross-modal transfer of response features specific to the interstimulus interval (ISI) between a conditioned stimulus (CS) and an unconditioned stimulus. Rabbits were given initial training with a stimulus (CSA) in one modality (e.g., tone) at a designated ISI (e.g., 600 ms). Training was then shifted to a new stimulus (CSB) in another modality (e.g., light) at a new ISI (e.g., 400 ms). The timing of early conditioned responses (CRs) to CSB reflected the ISI of CSA. Ultimately, CRs to CSB shifted to a temporal location conforming to the ISI of CSB. When the ISI of CSB was shorter than that of CSA, CRs to CSA also shifted to a locus conforming to the ISI of CSB. The present results confirmed previous findings that training in one CS modality accelerates CR acquisition to a CS in another modality. The findings are compared with the transfer of response patterns in instrumental learning sets and are discussed regarding their implications for theories of cross-modal transfer.     

Second-order conditioning of the rabbit's nictitating membrane response. Interstimulus interval and frequency of CS-CS pairings.

Second-order conditioning of the rabbit's nictitating membrane response (NMR) was investigated when second-order trials (CS1-CS2) were intermixed with first-order trials (CS2-US) from the outset of training. Experiment 1 showed that CR acquisition to CS1 was inversely related to the CS1-CS2 interval but nevertheless extended to an interval of 8,400 ms. Experiment 2 revealed that CR acquisition of CS1 was an inverted-U function of the number of CS1-CS2 trials relative to a fixed number of CS2-US trials. Experiment 3 directly contrasted second-order conditioning with a reinforced serial compound procedure (CS1-CS2-US) and a mixed procedure in which second-order trials were intermixed with the reinforced serial compound. Second-order conditioning was about half the strength of either the reinforced serial compound or the mixed procedure, which were similar. The present results are discussed with respect to the relative strength of excitatory and inhibitory processes in second-order conditioning.     

Second-order conditioning of the rabbit’s nictitating membrane response

Second-order conditioning of the rabbit’s nictitating membrane response (NMR) was investigated when second-order trials (CS1-CS2) were intermixed with first-order trials (CS2-US) from the outset of training. Experiment 1 showed that CR acquisition to CS1 was inversely related to the CS1-CS2 interval but nevertheless extended to an interval of 8,400 ms. Experiment 2 revealed that CR acquisition of CS1 was an inverted-U function of the number of CS1-CS2 trials relative to a fixed number of CS2-US trials. Experiment 3 directly contrasted second-order conditioning with a reinforced serial compound procedure (CS1-CS2-US) and a mixed procedure in which second-order trials were intermixed with the reinforced serial compound. Second-order conditioning was about half the strength of either the reinforced serial compound or the mixed procedure, which were similar. The present results are discussed with respect to the relative strength of excitatory and inhibitory processes in second-order conditioning.     

In the blink of an eye: Real-time stimulus factors in delay and trace conditioning of the rabbit's nictitating membrane response

Since Pavlov, theories of conditioning have assumed that CR evocation is governed by a series of internal stimuli generated by the CS. This hypothesis was tested in conditioning of the rabbit's nictitating membrane (NM) response by attempting to manipulate the internal sequence through truncating a delay CS and extending a trace CS on test trials. These perturbations of CS duration produced large deficits in CR likelihood and smaller alterations in the CR's time course. As predicted by many models of conditioning, the onset of the CS appeared to play a large but not exclusive role compared to CS duration and CS offset in both evocation and timing of the CR. The results are discussed with respect to their implications for real-time models of conditioning.     

Extinction revisited: similarities between extinction and reductions in US intensity in classical conditioning of the rabbit's nictitating membrane response

The mechanisms of extinction were examined by reducing the intensity of the unconditioned stimulus (US) after acquisition training to determine whether such reductions lie on a continuum with CS-alone extinction. The experiments revealed that reductions in US intensity yielded extinction-like effects. Specifically, there were proportional reductions in the daily mean level of responding across sessions. There were also persistent within-session declines and between-session increases of responding analogous to spontaneous recovery. Surprisingly, even when US intensity was held constant, within-session declines and between-session increases were apparent. The results are discussed with respect to possible contributions from unlearning, new learning, generalization decrement, and nonassociative loss, especially CS-specific attentional changes and CR-specific reactive inhibition.     

Temporal discrimination using different feature--target intervals in classical conditioning of the rabbit's nictitating membrane response

In a typical conditional discrimination, a target stimulus (X) is reinforced during one feature cue (A-->X+), but not during another feature cue (B-->X-). The present experiments used only a single "feature" cue (a 66-sec tone). On half of the trials, the target stimulus (a 400-msec light) was paired with the reinforcer when the feature-target interval was one duration (e.g., 5 sec). On the remaining trials, the interval was different (e.g., 45 sec), and the target stimulus was presented without the reinforcer. All the animals acquired this temporal discrimination, and subsequent testing with other feature-target intervals yielded generalization-like gradients. These results provide solid evidence that each portion of a feature cue is encoded in a distinctive fashion. Had temporal encoding not occurred, the feature cue would have been just as ambiguous a predictor of the reinforcer as was the target stimulus, and discrimination would not have been possible. The integration of real-time temporal encoding mechanisms into models of conditional discrimination is discussed.