Abstract
Performance on visual tasks can be improved via training or experience, and this is called visual perceptual learning (VPL). However, this improved performance is limited to the trained task’s specifics, i.e. when the spatial position of the stimulus is changed, the improvement disappears. Yet, recent research shows that variability along task-irrelevant stimulus dimensions can alter this characteristic. We argue that variability determines which neurons undergo plasticity in VPL, and depending on these neurons’ invariance properties, generalization or specificity is achieved. We trained two groups of participants with almost identical tasks, only changing which task-irrelevant dimension varied between trials. In particular, we created variability by randomizing spatial phase in one training group and contrast in the other. After training, we tested for transfer to a new spatial location in both groups. Phase-invariant neurons emerge later in the visual processing hierarchy, compared to contrast-invariant neurons (e.g., complex, and simple cells), and hence phase-invariant neurons have larger spatial receptive fields. Due to this, we hypothesized that varying the phase of the training stimuli over trials would give rise to generalization in space due to neurons which are phase invariant taking a role in the training. On the other hand, as contrast-invariant neurons appear earlier in the hierarchy, we expected the learning to be more specific when participants were trained with varying contrast. We found that the randomizing phase of the training stimulus resulted in complete generalization of the improvement to a new spatial location, contrary to randomizing contrast. Our results show that which neural populations undergo plasticity with VPL is determined by the training task demands, and in turn, this affects generalization and specificity of behavioral improvements.