Representation Learning Sample Clauses

Representation Learning. In this section, we seek an efficient and flexible representa- tion that is able to capture relevant structure of our navigation tasks. We build upon our previous work [17] from which we highlight the key aspects here, starting with the following definition, Definition 1 (Controller Graph (CG)). A weighted labeled graph G = (V, E, W ) with vertices V, edges E and a transi- tion matrix W , such that V ⊂ S and E = {(v , v ) | w > from earlier work of Xx and Xxxxxxx [16] in using IRL in infinite state spaces via sampled trajectories. 0, ∀vi, vj ∈ V, a ∈ A}. i j a i,j