Independently trained RNNs with non-similarity-based cue encoding exhibit idiosyncratic match error profiles.

(A) Match errors plot similar to Fig 3B: A in sequences shown at the top represents the cue of interest, and X represents any cue other than A, e.g. in the top row A = 1; dark blue bars indicate X-1-X-1 cue sequences and light blue bars indicate X-X-1-1 cue sequences. A identities for each row is given on the right. For RNN trained on 20k trials and exhibiting recency behavior, both dark and light bars show an even distribution of match errors across different lead cues. This parity across lead cues (all 25% of total errors (horizontal dotted line in each graph) could be suggestive of a true recency-based strategy. Error bars are 95% CI. (B) and (C): Match error plots for two independently trained networks (referred to as Network 1 and Network 2) after 20k training trials. Individual RNNs have widely varying match error profiles, indicating idiosyncratic underlying network state representations. (B) Same as (A) but match errors for an individual network, Network 1. (C) Same as (B) but for Network 2.