Learning produces a hippocampal cognitive map in the form of an orthogonalized state machine
Comments:
- It would be interesting to see a more fine-scale analysis of the links between neural activity and learning (and maybe with the models)
-
especially, at the individual mouse level, do the changes in the map predict or follow specific changes in learning? See for example the mouse that seemed to be an expert for 1 day and then reverted to a random strategy
.
- Mice learn the new cues faster compared to when they were naive, but without a control for familiarization with the environment, experimental procedures etc. it is unclear if this faster learning reflects habituation or generalization.
- Comparison of model ‘performance’ is very interesting but we are wondering how the intitial conditions and chosen parameters of each model influence the results. It is nice to see some discussion of this, but how could it be made more systematic?
- It would be interesting to test for signatures of temporal context, for example, does the activity orthogonalization remain regardless of time passed since experiencing the cue?
- Did any of the mice not learn? How did their activity differ from the expert mice?
- We believe the suggestion about the point clouds in the UMAP plot relating to replay should be more substantiated.