Amygdala neurons
Monkeys watch each other learning reward-predicting images for choosing between two rewards. A group of amygdala neurons signals which reward the observed monkey will be choosing (top right). Another group of amygdala neurons distinguishes which animal is making the choice (right). A third group of amygdala neurons takes these transitions from value coding to choice coding of the partner monkey (bottom right) in compliance with a formal choice model that incorporates these individual activities (bottom left).
Monkeys watch each other learning reward-predicting images for choosing between two rewards. A group of amygdala neurons signals which reward the observed monkey will be choosing (top right). Another group of amygdala neurons distinguishes which animal is making the choice (right). A third group of amygdala neurons takes these transitions from value coding to choice coding of the partner monkey (bottom right) in compliance with a formal choice model that incorporates these individual activities (bottom left).
Monkeys repeatedly choose between saving an earned juice reward for future consumption with interest and immediate spending the save reward. A group of amygdala neurons transitions from value coding to choice coding (left). Only a small number of amygdala neurons are necessary to make more than 90% accurate choice predictions (right). This is the kind of information the next postsynaptic neurons receive in order to execute the choice. Fine analysis reveals a gradual build-up of neuronal activity towards a spend choice (bottom). Thus, the neuronal activity tracks an internally controlled sequential choice progress in the absence of external cues, as it coccurs only in choice trials and not in externally instructed imperative trials and does not reflect reward expectation, sensory stimulation or action preparation.
Monkeys repeatedly choose between saving an earned juice reward for future consumption with interest and immediate spending the save reward. A group of amygdala neurons transitions from value coding to choice coding (left). Only a small number of amygdala neurons are necessary to make more than 90% accurate choice predictions (right). This is the kind of information the next postsynaptic neurons receive in order to execute the choice. Fine analysis reveals a gradual build-up of neuronal activity towards a spend choice (bottom). Thus, the neuronal activity tracks an internally controlled sequential choice progress in the absence of external cues, as it coccurs only in choice trials and not in externally instructed imperative trials and does not reflect reward expectation, sensory stimulation or action preparation.
The activity of some amygdala neurons ramps up to an expected singular reward (blue) but shows a more sustained elevation of activity when rewards are more spread out and the expectation lasts over a longer period (blue). Thus, the reward expectation in these neurons depends on the time of the reward, exactly what expectations should do.
The activity of some amygdala neurons ramps up to an expected singular reward (blue) but shows a more sustained elevation of activity when rewards are more spread out and the expectation lasts over a longer period (blue). Thus, the reward expectation in these neurons depends on the time of the reward, exactly what expectations should do.
Reward prediction (V) learning depends on the reward being contingent on the stimulus. A reward may occur also in the background without a stimulus (non-contingently; p(reward|stimulus = p(reward|no stimulus)), but this reward does not add any information to the stimulus and does not induce learning, even though the stimulus is paired with the reward. Amygdala neurons capture this fundamental requirement (bottom). Thus, stimulus-reward pairing is not enough for learning, although it appears so when pairing and contingency are not separated (top). This is what Pavlov did not know but Rescorla discovered (with punishers; 1967).
Reward prediction (V) learning depends on the reward being contingent on the stimulus. A reward may occur also in the background without a stimulus (non-contingently; p(reward|stimulus = p(reward|no stimulus)), but this reward does not add any information to the stimulus and does not induce learning, even though the stimulus is paired with the reward. Amygdala neurons capture this fundamental requirement (bottom). Thus, stimulus-reward pairing is not enough for learning, although it appears so when pairing and contingency are not separated (top). This is what Pavlov did not know but Rescorla discovered (with punishers; 1967).