Fast Perceptual Learning in Visual Hyperacuity

TLDR

The human visual system achieves hyperacuity in spatial discrimination tasks, such as determining the sign of a vernier offset, with precision finer than a photoreceptor’s diameter. The study proposes that this performance relies on a rapid, few‑example learning process occurring early in the visual pathway. The authors demonstrate that a simple network can be synthesized from a small set of task examples to reach the required hyperacuity performance. Psychophysical experiments confirm the model’s predictions, showing fast stimulus‑specific learning in humans that does not transfer between slightly different hyperacuity tasks.

Abstract

In many different spatial discrimination tasks, such as in determining the sign of the offset in a vernier stimulus, the human visual system exhibits hyperacuity by evaluating spatial relations with the precision of a fraction of a photoreceptor's diameter. It is proposed that this impressive performance depends in part on a fast learning process that uses relatively few examples and that occurs at an early processing stage in the visual pathway. This hypothesis is given support by the demonstration that it is possible to synthesize, from a small number of examples of a given task, a simple network that attains the required performance level. Psychophysical experiments agree with some of the key predictions of the model. In particular, fast stimulus-specific learning is found to take place in the human visual system, and this learning does not transfer between two slightly different hyperacuity tasks.