Learned object-specific invariances are revealed by the effect of intensity on environmental sound recognition
Year: 2018
Type: preprint
Abstract: The central challenge of recognition is that similar entities in the world do not always produce similar sensory signals. One solution to this challenge could be to generically remove or separate irrelevant sources of variation from object representations. Alternatively, robust recognition might entail learning the extent and type of variation encountered for particular object classes. The effect of sound intensity on auditory recognition provides a simple test of the two hypotheses for real-world sounds, because canonically quiet sounds (e.g. salt-shakers) are not normally encountered with high intensities at the ear. We measured the effect of intensity on the recognition of common environmental sounds. For canonically loud sounds (e.g. jackhammers), recognition improved with intensity. But for canonically quiet sounds, recognition peaked at moderate intensities and declined thereafter. These results suggest that even basic forms of invariance are at least in part learned individually for object classes rather than instantiated generically.
Authors: Sam Norman-Haignere, Josh Mcdermott
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W3164718596