Search Papers | Poster Sessions | All Posters
Poster B30 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Attention-Dependent Perceptual Learning in a CNN Model of the Visual System
Thomas Maher1, Grace Lindsay1; 1New York University
Perceptual learning is defined by increased performance on a perceptual task following prolonged practice. Many studies have observed that rates of perceptual learning can depend on the strength of visual attention deployed during the learning. Despite attempts to relate perceptual learning and attention, the exact mechanism behind this relationship remains unknown. Here, we propose a con- volutional neural network (CNN) model of visual percep- tual learning for the purpose of elucidating this relation- ship. Our model uses an attention system along with a lo- cal learning rule that, through weight updates, solidifies the impact of attention. We found that the model’s per- formance on a precise visual task increased as a result of the local learning rule and that this effect was depen- dent on the magnitude of the attention modulation. This suggests that modulatory attention and plasticity in early visual areas are sufficient for inducing perceptual learn- ing.
Keywords: Attention Perceptual Learning CNN