Fig. 2: Diverse asymmetric scaling of RPEs predicts optimism.
From: Distributional reinforcement learning in prefrontal cortex
a, An example neuron’s responses at each of the task epochs: choice, feedback on rewarded trials and feedback on unrewarded trials. β+ and β− are betas corresponding to the scaling of positive and negative RPEs. Betas are calculated on the mean firing rate in a 200- to 600-ms window after feedback. Error bars denote s.e.m. Note that, for rewarded and unrewarded trials, we do not display the lowest and highest value levels, respectively, owing to a small number of trials giving unreliable traces. b, Histogram showing a diversity of asymmetric scaling across ACC RPE neurons. Coloring denotes optimism as defined by asymmetric scaling, with red more optimistic. c, Same format as Fig. 1d but for asymmetric scaling consistency: mean R = 0.32, P = 0.014 across data partitions. Each point denotes a neuron. d, Asymmetric scaling estimated at feedback predicted reversal point at choice: R = 0.41, P = 0.0079. Each point denotes a neuron.
