Emergence of belief-like representations through reinforcement learning — part 7/7

Cleaned-text code points [87415, 88565). Bibliography omitted here; source.md includes it.

Fig S3. Value RNNs trained on the Babayan task exhibit one fixed point. a. RNN
activity during two example trials, one during Block 1 (left) and the other during Block
2 (right). Same as Fig 7D. Here we also include RNN activity trajectories if each reward
had been omitted. While activity for the Block 2 trial initially returns to the putative ITI 2
state, it eventually returns to the true fixed point at ITI 1 b. Distance of RNN activity
from the single fixed point (e.g., ITI 1 in panel a) following an odor observation (i.e., an
omission trial). While the maximum ITI duration is theoretically infinite, the maximum
ITI duration in the training data was at t = 65. RNN activity on Block 2 trials therefore
remained separate from the activity on Block 1 trials for the range of experienced ITI
durations.
S3
.CC-BY 4.0 International licenseavailable under a 
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made 
The copyright holder for this preprint (whichthis version posted April 7, 2023. ; https://doi.org/10.1101/2023.04.04.535512doi: bioRxiv preprint