Learning in rich networks involves both positive and negative associations.

Abstract

Adaptive behaviors are believed to be shaped by both positive (the strengthening of correct associations) and negative (the pruning of incorrect associations or the building of inhibitory associations) forms of associative learning. However, there has been little direct documentation of how these basic processes participate in the learning of rich associative networks that support cognitive behaviors like categorization. Although negative associative learning is an important component of theories of development, it is not clear whether it involves acquiring specific (experience-dependent) content or represents a more general aspect of (experience-expectant) development. The authors thus trained pigeons on a complex many-to-many learning paradigm previously established as an analog to human word learning. Pigeons learned to map 16 objects onto 16 distinct report tokens; the authors manipulated the amount of negative associative learning that could occur by restricting which tokens were available as incorrect options. In testing, accuracy was lower on trials with foils that had not been presented with a target than on trials with previously experienced foils. Moreover, when the correct token was withheld, pigeons preferred foils novel to the target object over previously experienced foils. A second experiment replicated these results and further found that these effects only emerged after some positive associations had been acquired. Findings indicate that the learning of rich associative networks does not depend solely on positive associative learning, but also on negative associative learning; this conclusion has important implications for basic learning theories in both animals and humans, as well as for theories of development. (PsycINFO Database Record

Feature predictiveness and selective attention in pigeons’ categorization learning.

“…A process similar to that reported by Roembke et al (2016) could underlie pigeons’ observing responses in our study. The birds may have learned that the irrelevant features did not allow them to predict the correct response.…”

“…Roembke, Wasserman, and McMurray (2016) recently trained pigeons to associate each of 16 object images with one particular report image or ‘pexigram.’ On each trial, two pexigrams were presented: one correct (a pexigram that was always presented with a particular object) and one incorrect (a randomly selected pexigram). For a given target object, only 8 pexigrams were available as potential report options (the correct pexigram and 7 possible foils); the other 8 foil pexigrams were withheld for testing.…”

“…In testing, when Roembke et al (2016) presented each of the objects along with the correct pexigram and, as the alternative pexigram, a foil that had never been presented with that object (but had been presented with other objects) during training, pigeons’ accuracy decreased, even when the correct pexigram was still available. These results suggest that pigeons were learning not only a positive association between a specific object and the correct pexigram, but also negative (or inhibitory) associations between a specific object and the incorrect pexigrams.…”

Poor numerical performance of guppies tested in a Skinner box

“…Therefore, the fish can solve the task by recognising one of the two stimuli or even a portion of a stimulus. The simple rule "approach the stimulus if it is the reinforced one and avoid it if is the non-reinforced one" 40,41 should favour a habit and allow learning colour and shape discriminations in a Skinner-box condition. Conversely, in a numerosity discrimination task, each trial has a different pair of stimuli, because position, size, and density of the items vary systematically.…”

Simultaneous training on overlapping grapheme phoneme correspondences augments learning and retention

“…yielding an opportunity to tune multiple associations on any given trial (Roembke, Wasserman, & McMurray, 2016…”