Figure 5
From: Neural network based successor representations to form cognitive maps of space and language
Reinforcement learning to navigate a spatial environment: We reproduced the rat maze experiment published by Alvernhe et al.36. (left column). Therefore, we simulated the corresponding maze environment, small green squares indicate three different sample starting positions (second column). Based on the transition probabilities to neighboring states we calculated the successor representation of the maze as ground truth (third column). The predicted SR of the trained network, i.e. the firing patterns of the artificial place cells are very similar to the underlying ground truth (right column).
