Figure 5:

Learned mosaics exhibit a phase transition as a function of input and output noise. (A-C) Spatial kernel centers for Slow (A), FastA (B), and FastB (C) as a function of ${\sigma }_{\text{in}}$ and ${\sigma }_{\text{out}}$. In all three cases, the optimal configuration changes from aligned to anti-aligned when output noise increases or input noise decreases. Blue bars denote alignment as measured by median distance to the nearest RF center of the opposite type.