A neural model for compass neurons

Fig. 9:

Schematic representation of the neural network: The input layer contains three input-neurons representing the POL-neurons. In the successive layers, a larger group of higher-order neurons with more closely spaced tuning e-vectors are built by connecting them to different pairs of neurons in the previous layers, providing the system with useful resolution. The output-layer consists of an array of narrowly-tuned "compass neurons" each representing a certain body orientation with respect to the zenithal e-vector. Each body orientation produces a characteristic activity pattern among the compass neurons with only a small, closed group of neurons active while the other neurons are inactive (see histogram at lower left).