Fig. 2

Learning towards fixed and switching goals. a Categorical accuracy against the goal of learning. Blue indicates the MVG of Airplane and Truck. Blue line is a single line with high frequency. Y-axis is the accuracy of the learning goal. Black connects the epochs needed to learn Airplane in the MVG. Red indicates the FG of Airplane. Purple indicates Rcost. b, c Categorical accuracy against unlearned goals Cat and Ship. The weights of the initial neural network are from the last epoch of (a). e, d Pearson’s correlation coefficient (cc) against before and after learning unlearned goals Cat and Ship. The cc of the filter elements of the initial neural network and the last neural network in (b, c) are shown. Blue indicates the cc of all the filter elements. Red indicates the cc of elements in the filter corresponding to intra-module links. Purple indicates the cc of elements in the filter corresponding to inter-module links