Figure 3

Learned successor representation (SR) matrices and corresponding ground truths. Learned SR matrices (left column) are very similar to their corresponding ground truth SR matrices (right column). (A) For a discount factor of \(\gamma =0.3\), the root-mean-squared error (RMSE) between learned and ground truth SR matrix is 0.02034. (B) For \(\gamma =0.7\), the RSME is 0.01496. (C) For \(\gamma =1.0\), the RMSE is 0.00854. Note that, x-axes denote index of starting state, y-axes denote index of successor state.