Please use this topic to ask questions about this session.
I would like to answer some of the questions:
“Why there is no difference between ventarl and dorsal parts, i.e. Nucleus Acumbens, VTA and dorsal striatum, SNc?”
In the DopAct framework the Valuation, Goal-directed and Habit systems are mapped on Ventral, Dorsomedial and Dorsolateral striatum respectively. It is assumed that that dopaminergic neurons in Valuation and Habit systems are mapped on VTA and SNc respectively, and dopaminergic neurons in the Goal-directed system may be in both these areas. In the full paper (https://elifesciences.org/articles/53262) the relationship of the model to experimental data from these regions are discussed.
Let me answer another question:
“O1: Action planning is basically generation of more predictions due to the triggered PE.”
Indeed this is a nice way of putting the key idea of the model.
“What is the basis of generating these actions ? As in are they just a result of parameter tweaking (in a variational sense) or some cached actions for the best probability of reward ?”
I did not describe it in sufficient detail during the talk, but you can think about habit system as indeed storing “cashed actions”, while the goal-directed system, is optimizing the action to minimize its prediction error. For more details, see the full paper (https://elifesciences.org/articles/53262).