Abstract
Stimulus-specific perceptual learning (PL) often has been found in relatively simple tasks that require observers to discriminate stimuli that vary along a single dimension. Such learning has been interpreted as showing that training alters the response properties of mechanisms in early visual cortical areas. Stimulus-specific PL also has been found in situations that are arguably more complex, namely identification tasks using filtered random textures that varied on multiple dimensions: PL does not transfer to contrast-reversed versions of the training textures or to textures rotated by 180 deg [Hussain et al., J Vis, 2009, 9(4):20]. Such findings suggest that PL may alter the response properties of mechanisms in higher-level visual areas such as inferotemporal cortex. One potential criticism of the Hussain et al. findings is that stimulus specificity was assessed using transformations that altered the spatial distribution of the black and white blobs in the textures, and essentially created a new set of stimuli. According to this argument, PL in a texture identification task may generalize to textures presented in different spatial locations, without altering the spatial distributions of features. We tested this idea by measuring accuracy in a 1-of-5 texture identification task. Practice on Day 1 consisted of interleaved trials with two different sets of textures: one set was presented slightly above the fixation point and the other was presented below the fixation point. On Day 2, subjects continued the task either with stimuli in the same positions seen during Day 1 or with the top-bottom positions switched. We found evidence for both generalized and stimulus-specific learning: presenting textures in new positions on Day 2 caused accuracy to decrease significantly, but not to levels seen at the beginning of Day 1. Hence, a component of learning in this relatively complex texture identification task is stimulus specific.
Meeting abstract presented at VSS 2015