Brown University researchers have developed a neural-network model that can be deceived by optical illusions.
The researchers started with a computational model limited by anatomical and neurophysiological data of the visual cortex, designed to measure how neighboring cortical neurons send messages to each other and adjust their responses when confronted with stimuli such as contextual optical illusions. The model included a pattern of hypothesized feedback connections between neurons that can excite or inhibit a central neuron's response, depending on the visual context.
After assembling this model, the team presented various context-dependent illusions, "tuning" the strength of the feedback excitatory or inhibitory links so neurons responded in a manner consistent with neurophysiology data from the primate visual cortex.
Tests on assorted contextual illusions again indicated the model perceived the illusions like a human.
The team aims to improve contextual awareness of artificial-vision algorithms by adding horizontal connections tuned by context-dependent optical illusions.
From News from Brown
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found