Closed rcoreilly closed 10 months ago
A simple way of ensuring at least some learning at time of US is to add a USTrace learning param:
tr = USTrace * (input now at US) + (1-USTrace) * syn.Tr
should help make BLA learning more robust to changes over goal pursuit -- potentially useful regardless of #281 changes.
This works in initial eboa testing to allow curiosity-driven learning to acquire CS associations.
A simple way of ensuring at least some learning at time of US is to add a USTrace learning param:
should help make BLA learning more robust to changes over goal pursuit -- potentially useful regardless of #281 changes.