Associative Learning Through Acquired Salience

Abstract

Most associative learning studies describe the salience of stimuli as a fixed learning-rate parameter. Presumptive saliency signals, however, have also been linked to motivational and attentional processes. An interesting possibility, therefore, is that discriminative stimuli could also acquire salience as they become powerful predictors of outcomes. To explore this idea, we first characterized and extracted the learning curves from mice trained with discriminative images offering varying degrees of structural similarity. Next, we fitted a linear model of associative learning coupled to a series of mathematical representations for stimulus salience. We found that the best prediction, from the set of tested models, was one in which the visual salience depended on stimulus similarity and a non-linear function of the associative strength. Therefore, these analytic results support the idea that the net salience of a stimulus depends both on the items' effective salience and the motivational state of the subject that learns about it. Moreover, this dual salience model can explain why learning about a stimulus not only depends on the effective salience during acquisition but also on the specific learning trajectory that was used to reach this state. Our mathematical description could be instrumental for understanding aberrant salience acquisition under stressful situations and in neuropsychiatric disorders like schizophrenia, obsessive-compulsive disorder, and addiction.

Noise Improves Visual Motion Discrimination via a Stochastic Resonance-Like Phenomenon

“…We employed standard non-linear programming techniques to fit the experimental data to both of these models (Treviño, 2015). …”

Adrenergic Modulation of Visually-Guided Behavior

“…Reaction time (RT) distributions constitute a rich source of information to understand perceptual processes. Factors such as stimulus saliency and the uncertainty of the responses influence the shape of these distributions (Treviño, 2016 ). Besides, there is ample evidence showing how the skewed shape of RT distributions depends on task difficulty and the rate at which information becomes available to solve it (Smith and Ratcliff, 2004 ).…”

“…We derived four main models from this equation: (1) No gain modulation (Model 1: d x = 0, d y = 0; free parameters: K = 2); (2) Input gain modulation (Model 2: d x = variable, d y = 0; K = 3); (3) Output gain modulation (Model 3: d x = 0, d y = variable; K = 3); and (4) Input/Output gain modulation (Model 4: d x = variable, d y = variable; K = 4). We used the Akaike Information Criterion (AIC) to identify the best predictive model (Treviño, 2016 ). Briefly, the second order AIC ( AIC C ) compensates for sample size by increasing the relative penalty for fits with small data sets:…”

Multimetric behavioral profile and psychological traits

“…Human learning flexibility can be integrated into a general framework in which model parameters represent and lead to variance in learning (Rescorla and Wagner, 1972;Trevino et al, 2011;Byrom, 2013;Glautier, 2013;Trevino, 2015;2020).…”