Learning curves for different read-out configurations.

The read-out network in the main text consists of two separate networks, which are not interconnected. A. Cartoon of read-out network without recurrent excitatory connections. B-D: The learning curves when the recurrent connections in the two separate motif networks are zero. The same relearning protocol as in Fig 3 and S4 Fig is used. E. Cartoon of read-out network when the two motif networks are combined and interconnected into one network. F-H: The learning curves when the two motif networks are combined and interconnected into one network. In this case, the same connections as listed in Table 2 are used but multiplied by , and , . The sparsity of the connections remains p = 0.2. The same relearning protocol as in Fig 3 and S4 Fig is used.

(TIF)