Fig. 6: Location transfer results.
From: Visual perceptual learning of feature conjunctions leverages non-linear mixed selectivity
a Schematic illustration of the two transfer locations, near and far (3 and 6 dva, respectively, from the training location). b Differential receptive field (RF) sizes of pure orientation and color selectivity neurons would lead to different transfer results for the two features: larger color RFs entail larger overlap between trained and transfer locations (area marked in orange), and hence larger TIs compared to orientation. In contrast, mixed selectivity neurons have identical RFs for color and orientation. This predicts no difference in Tis between the two features. Red circles: trained location RFs, yellow circles: transfer locations RFs. c Transfer Indices (TIs) for conjunction discrimination. There was no significant difference between locations (near and far) and experiments (Location: F(1,33) = 1.6076, p = 0.2137, η2 = 0.0465; Experiment: F(1,33) = 0.1302, p = 0.7206, η2 = 0.0039; Location × Experiment: F(1,33) = 1.2071, p = 0.2798, η2 = 0.0353). d TIs for single features. Orientation and color TIs are significantly different from each other (Experiment 1, near TIs: p = 0.002, d = 1.1, far TIs: p = 0.1641, d = 0.73; Experiment 2, near TIs: p = 0.0547, d = 1.13, far TIs: p = 0.0273, d = 1.1232, Experiment 3, p = 0.0195, d = 1.0731). Dashed black lines in panels b and c show full transfer (TI = 1). CLGF Conjunction Learning with Global Feedback, CLFF Conjunction Learning with Feature-specific Feedback, SFL Separate Feature Learning, OL Orientation Learning. Error bars represent standard deviation. ** denotes p < 0.01, * denotes p < 0.05, n.s. denotes non-significant.
