Fig. 4: Improved re-learning is not caused by a biased starting point.

Exponential and power law models in Fig. 3 were fit to the entire re-learning period in exposure 2. No-feedback washout group started this period at a biased state. Inset (a) (‘unadjusted’) provides the terminal washout state (average angle on last four washout cycles (control shows average on last four baseline cycles). We refit the exponential model to participant data after removing the initial cycles in the re-learning phase required for Control, Abrupt washout, and Gradual washout groups to ‘catch up’ to the no-feedback washout starting point. Inset (d) shows the initial reaching angle on the adjusted starting cycle. The exponential model fits to the adjusted data are shown in Inset (a). Learning curves in (a) are overlaid with the control group for ease of comparison. The exponential model fit is shown overlaid on the measured learning curve. Inset (b) shows the associated rate parameters for the exponential model. Inset (e) provides a comparison between re-learning rates (or the naïve learning rate in the control group) and the adaptation starting point (i.e., the terminal washout state). x-Axis provides the terminal washout state for each participant (the collapsed individual points in c). y-Axis provides the learning rates shown in (b). Both axes were zero-centered at the group level (i.e., the group mean was subtracted from each data point prior to visualization and analysis). Statistics in (e) refer to a linear regression. Error bars in all insets indicate mean ± SEM. Single dots in (b–e) are individual subjects. Statistics: ^*p < 0.05, ^**p < 0.01, and ^***p < 0.001.