Thread by @vinnycosta_phd

!!! New paper now out in @NeuronCellPress !!!

We show that neurons in the amygdala and ventral striatum represent different computations when monkeys decide to explore a new opportunity or exploit what they already know

cell.com/neuron/fulltex…

One of the more challenging problems in reinforcement learning is referred to as the explore-exploit dilemma. It refers to the tradeoff we face when we have to decide whether to repeat a past choice or explore a new opportunity.

In noisy environments where you can never be especially certain about the value of anything, we know from seminal studies by @nathanieldaw, @BecketEbs, and others that prefrontal cortex is important for switching between states of exploration and exploitation.

We took a different approach & induced explore-exploit tradeoffs by introducing novel options amidst choice options the monkeys had already learned about.

We know motivational circuits facilitate RL, do they also support computations relevant for deciding when to explore?

The short answer is yes! When we use a partially observable Markov decision process (POMDP) model to break down the value of the monkeys' choices we find different encoding schemes in the amygdala and ventral striatum

When the monkeys choose to explore novelty, the bonus in value associated with exploring and finding out that there is a better option to choose from is more strongly encoded in the amygdala compared to the ventral striatum.

Whereas the immediate expected value of the option, based on the past outcome and choices, is more strongly encoded in the ventral striatum compared to the amygdala.

Therefore at the population level neurons in the amygdala and ventral striatum, simultaneously encode the exploration and exploitation value of choices.

We think this is exciting because it raises some new questions, in a computationally tractable framework, about how motivational neural circuits might contribute to impulsive and compulsive forms of exploratory learning and decision-making.