Figure 4.

Schematic of a basis-function network to achieve a coordinate transformation. For simplicity, stimulus representation and eye position are encoded in a single dimension each represented by a population code. In this case, a diagonal readout of this layer yields a subtraction of the two values encoded in the input layers—a head-centred representation. (a) The eye is looking straight ahead at 0°. The stimulus is at 0° in the world and therefore also at 0° on the retina. (b) The eye has moved to +5°. Since the stimulus is still at 0° in the world, it appears at −5° on the retina. The network computes the correct cranio-topic stimulus position in both cases.