Performance of model-based algorithms on a 10-item list-linking procedure

During training, two 5-item lists were trained (adjacent pairs only). In the Linking condition, actors received training on the linking pair EF. In the No Linking condition, no extra training on EF was provided. Actors were then tested on the 45 possible pairings in the resulting 10-item list. (A) Mean performance of RL-Elo on non-terminal pairs throughout training and testing, sorted by symbolic distance. After training, dashed lines represent the Linking condition, while solid lines represent the No Linking condition (B). Mean performance of RL-Elo for each of the 45 pairs at transfer, sorted by symbolic distance. Light-colored points represent the Linking condition, while dark-colored points represent the No Linking condition. Critical pairs are shaded in gray. (C) Mean performance of RL-Elo for each of the 45 pairs at the end of training. (D) to (F) As above, but reporting simulated performance of the Sequential Monte Carlo (SMC) algorithm. (G) to (I) As above, but reporting the simulated performance of Betasort