Learning to generalize stimulus-specific learning across contexts

Perceptual learning (PL) in a texture identification task occurs because observers become more sensitive to diagnostic stimulus components, but the particular components that are learned vary across observers. Here, we encouraged observers to adopt a specific processing strategy by manipulating the diagnostic orientation structure of textures. Six targets were generated by applying a 0±30 deg orientation-filter to band-pass filtered white noise. Targets were embedded in non-diagnostic contexts created by filtering novel textures with a 90±30 deg orientation band. The remaining empty orientations served as potential cues for distinguishing the target and context. In a 1-of-6 identification task, observers were trained with the horizontal Target alone (hT-only) or in a vertical Context (hT+vC). Before and after training, observers were tested with hT-only and hT+vC stimuli and controls (vT-only and vT+hC). Training with hT-only produced strong stimulus- and context-specific learning; hT+vC training led to weak but stimulus-specific and context-generalizable learning. Next, we examined if context-generalization reflected a broader, stimulus-transferable learning by testing hT+vC trained observers with novel hT textures. Again, learning was minimal: almost half of the observers did not exhibit learning. For those who learned, we found no evidence of context-generalization or stimulus-transfer. Finally, to make it easier to distinguish targets and context during training, target contrast was held constant while context contrast was varied with a 1up/1down staircase. Significant learning occurred: across training blocks, observers tolerated increasingly higher levels of context contrast. Unlike previous experiments, training improved performance for both the same hT+vC and hT-only stimuli; however, training did not affect performance with novel hT or vT stimuli. Thus, learning was context-generalizable but not stimulus-transferable. In conclusion, learning to identify structure in a specific orientation is difficult when that information is embedded in non-informative orientation information, but this difficulty can be overcome by providing cues to distinguish components. Meeting abstract presented at VSS 2016

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.