Organizing Principles of Learning and Decision Making
Reinforcement Learning Meets Dopamine
Reinforcement learning theories suggest that agents are more likely to repeat actions that are followed by rewards. Dopamine -a neuromodulator molecule in the brain- is critical for such learning processes. This hypothesis is one of the most elegant success stories of systems neuroscience that relates specific brain mechanisms to behavioral psychology and computer science. Dopamine cells in the midbrain projecting to striatum -a subcortical structure serving as the input to basal ganglia- fire bursts of action potentials in response to stimuli that predict future rewards (Fig. 1). Multiple structures (cortex, hippocampus, amygdala etc) transfer representations of actions and stimuli to striatum using glutamate - an excitatory neurotransmitter. Coincidence of dopamine burst and glutamatergic inputs will lead to structural plasticity in the synaptic connections between striatal neurons and their inputs (Fig. 2).