Learning complex cell units from simulated prenatal retinal waves with slow feature analysis

Many properties of the developing visual system are struc-tured and organized before the onset of vision. Spontane-ous neural activity, which spreads in waves across theretina, has been suggested to play a major role in theseprenatal structuring processes [1]. Recently, it has beenshown that when employing an efficient coding strategy,such as sparse coding, these retinal activity patterns lead tobasis functions that resemble optimal stimuli of simplecells in V1 [2].Here we present the results of applying a coding strategythat optimizes for temporal slowness, namely Slow Fea-ture Analysis (SFA) [3], to a biologically plausible modelof retinal waves [4] (see figure 1). We also tested otherwave-like inputs (sinusoidal waves, moving Gauss blobs)that allow for an analytical understanding of the results.Previously, SFA has been successfully applied in modelingparts of the visual system, most notably in reproducing arich set of complex cell features by training SFA with nat-ural image sequences [5]. In this work, we were able toobtain complex-cell like receptive fields in all input con-ditions, as displayed in figure 2.Our results support the idea that retinal waves share rele-vant temporal and spatial properties with natural images.Hence, retinal waves seem suitable training stimuli tolearn invariances and thereby shape the developing earlyvisual system so that it is best prepared for coding inputfrom the natural world.