Fig. 6: The absence of blocking does not imply the absence of reinforcement prediction errors.
From: Learning with reinforcement prediction errors in a model of the Drosophila mushroom body
a Schematic of the protocol used to simulate blocking experiments. b Schematic of KC responses to the two conditioned stimuli, X and Y, when presented alone or as a compound. c, d Reinforcement predictions (RPs) for the two stimuli, averaged over the two test trials. Bars and whiskers: mean ± standard deviation. Circles: RPs from individual simulation runs, n = 50. Source data provided in Source data file. c Stimuli elicit independent KC responses during compound training. d Y gains a positive RP when KC responses to each stimulus are corrupted (\({p}_{{\rm{cor}}}^{{\rm{X}}}\,=\,0.8\), \({p}_{{\rm{cor}}}^{{\rm{Y}}}\,=\,0.2\)). e Performance indices during the test phase. When stimuli elicit independent KC responses (ind.), the weak RP for Y results in little choice preference for Y over the null option. When KC responses are corrupted (non-ind.), the positive RP for Y results in a relative strong preference for Y over the null option. Bars and whiskers: mean ± standard deviation, n = 20. Circles: PIs from batches of 50 simulation runs. Source data provided in Source data file. f Gradations in the blocking effect result from varying degrees of corruptions to X or Y during the compound stimulus training phase.
