Summary
Stimulus and location specificity are long considered hallmarks of visual perceptual learning. This renders visual perceptual learning distinct from other forms of learning, where generalization can be more easily attained, and unsuitable for practical applications, where generalization is key. Based on hypotheses derived from the structure of the visual system, we test here whether stimulus variability can unlock generalization in perceptual learning. We train subjects in orientation discrimination, while we vary the amount of variability in a task-irrelevant feature, spatial frequency. We find that independently of task difficulty, this manipulation enables generalization of learning to new stimuli and locations, while not negatively affecting the overall amount of learning on the task. We then use deep neural networks to investigate how variability unlocks generalization. We find that networks develop invariance to the task-irrelevant feature when trained with variable inputs. The degree of learned invariance strongly predicts generalization. A reliance on invariant representations can explain variability-induced generalization in visual perceptual learning, suggests new targets for understanding the neural basis of perceptual learning in high-order visual cortex, and presents an easy to implement modification of common training paradigms that may benefit practical applications.
Competing Interest Statement
The funders had no role in study design, data collection and interpretation, decision to publish, or preparation of the manuscript. ASD is a founder of Neuro-Inspired Vision and a member of its scientific advisory board. GLM and CMS declare no competing financial interests.
Footnotes
↵† Lead contact
This revision includes additional data, additional analyses, more details on some of the design choices, as well as an expansion of the discussion section. E.g., we include now an entire new control experiment that fully replicates our original finding with a different range of spatial frequencies and rules out alternative interpretations based on the idea of differential 'bracketing' of the transfer spatial frequency in the different training conditions. We have also added pretraining data on the different locations we tested that fully support our conclusions. And we included a control analysis for stimulus-specific adaptation, which we can now rule out as an underlying factor. Most importantly, despite this profound scrutiny, our results all remain the same.