Figure 1
From: Neurally plausible mechanisms for learning selective and invariant representations
Invariance and selectivity: (\(\boldsymbol{a_{1}}\), \(\boldsymbol{a_{2}}\), \(\boldsymbol{a_{3}}\)) intra- and inter-orbits log distances statistic for toy data (cyclic (\(a_{1}\)) and dihedral group (\(a_{2}\))) and rotated natural patches (\(a_{3}\)); (\(\boldsymbol{b_{1}}\), \(\boldsymbol{b_{2}}\), \(\boldsymbol{b_{3}}\)) selectivity in terms of cosine similarity for the three datasets with respect to the number of Heaviside nonlinearities