Mechanisms of offline motor learning at a microscale of seconds in large-scale crowdsourced data

Abstract

Performance improvements during early human motor skill learning are suggested to be driven by short periods of rest during practice, at the scale of seconds. To reveal the unknown mechanisms behind these "micro-offline" gains, we leveraged the sampling power offered by online crowdsourcing (cumulative N over all experiments = 951). First, we replicated the original in-lab findings, demonstrating generalizability to subjects learning the task in their daily living environment (N = 389). Second, we show that offline improvements during rest are equivalent when significantly shortening practice period duration, thus confirming that they are not a result of recovery from performance fatigue (N = 118). Third, retroactive interference immediately after each practice period reduced the learning rate relative to interference after passage of time (N = 373), indicating stabilization of the motor memory at a microscale of several seconds. Finally, we show that random termination of practice periods did not impact offline gains, ruling out a contribution of predictive motor slowing (N = 71). Altogether, these results demonstrate that micro-offline gains indicate rapid, within-seconds consolidation accounting for early skill learning.

Keywords: Consolidation; Human behaviour.

Fig. 1. Experiment 1: early learning of a new skill occurs largely offline.

Replication in a crowdsourced sample of 389 participants. a–d in-lab experiment (N = 27, 17 female, mean ± s.e.m. age 26.3 ± 0.83), e–h crowdsourced experiment (224 female, 39.6 ± 0.56) a, e Task: participants learned a motor skill task^,, over 36 trials (inset shows a single trial) consisting of alternating practice and rest periods of 10 s duration for a total of 12 min. b, f Skill was measured as the average inter-tap interval within correct sequences (tapping speed measured in keypresses/s)^, and is shown over the first 11 trials for the in-lab (b) and crowdsourced (f) group (see Supplementary Fig. 1a, c for all 36 trials). Micro-online changes were calculated as the difference in tapping speed (keypresses/s) of the first and last correct sequence within a practice period (blue in inset) and micro-offline changes as the difference between the last correct sequence within a practice period compared with the first of the next practice period (red in inset). The average number of correct sequences per trial is shown as green dots. c, g Trial-wise early learning (trials 1–5) composed of micro-offline (red), micro-online (blue), and total (black) performance changes (mean + s.e.m.). Note the presence of large micro-offline gains and total early learning in the initial trials in the absence of micro-online performance decrements. d, h Data points in the violin plot depict the sum of changes in performance over early learning trials in each participant, the red line denotes the mean (***P < 0.001).

Fig. 2. Experiment 2: stabilization of motor skill during short periods of rest.

a Task: learning of the target sequence was interfered by learning of another sequence either immediately (early interference, N = 118, first row), or 10 s after (late interference, N = 126, second row) each practice period. To avoid proactive interference of late interference on the following trial, a rest period of 10 s was introduced and the rest period in the early interference group matched to 20 s. In a control group, no interference was given, and each practice period was followed by 30 s of rest (N = 129). Training consisted of 12 trials amounting to 8 min. b Skill was measured as the average inter-tap interval within correct sequences (tapping speed measured in keypresses/s)^, and c as the number of correct sequences^,,,,. The performance curve of each group (red: early interference, magenta: late interference, cyan: no interference; mean + s.e.m.) is overlaid with the average of modeled performance curves. Note the shallower rise of the performance curve in the early as opposed to late interference group. d Model parameters (initial performance, maximum performance, and learning rate) of the learning curve for each experimental group, the line denotes the mean. Modeling of the number of correct sequences revealed a significant group difference in the learning rate, with early interference showing a shallower curve (P < 0.05, orange arrow). *P < 0.05, ***P < 0.001.

Fig. 3. Experiment 3: training under reduced practice period duration shows comparable micro-offline gains.

a Task: participants learned the motor skill task^,, over 48 trials (inset shows a single trial) consisting of alternating 5 s practice and 10 s rest periods for a total of 12 min, matched to the original experiment duration (Fig. 1). b Skill was measured as the average inter-tap interval within correct sequences (tapping speed measured in keypresses/s)^,. The group average performance curve is given in magenta (N = 118) and the group average of the 10 s practice period group (Experiment 1, Fig. 1) displayed for comparison in black (N = 212). c Trial-wise early learning. Each line depicts performance changes (micro-offline in red, micro-online in blue, total in black) per trial (mean + s.e.m.). Even under reduced practice period duration, total learning was closely accounted for by micro-offline gains (black and red lines) whereas micro-online performance changes fluctuate around 0 (blue line). Micro-offline learning remains high when halving practice period duration. d Data points in the violin plot depict the sum of changes in performance over early learning trials in each participant, the red line denotes the mean. In both groups (10 and 5 s), total early learning is accounted for by performance improvements during rest periods, but not during practice periods. ***P < 0.001. Note that participants with a high performance were selected in both groups due to required at least two correct sequences in each trial for calculation of microscale learning (“Methods”, “Data Analysis” section).

Fig. 4. Experiment 4: training under unpredictable-practice period duration shows comparable micro-offline gains.

Rhythmicity of practice-rest alterations may lead to preemptive slowing towards the end of each practice period that may contribute to micro-offline gains. Unpredictable-practice period durations (random 5, 6, 7, 8, 9, 10 s) prevent preemptive slowing. Task: participants learned the motor skill task^,, over 41 trials (inset shows a single trial) consisting of alternating 5–10 s practice and 10 s rest periods for a total of 12 min, matched to the original experiment duration (Fig. 1). b Skill was measured as the average inter-tap interval within correct sequences (tapping speed measured in keypresses/s)^,. The group average performance curve is given in magenta (N = 71) and the group average of the 10 s practice period group (Experiment 1, Fig. 1) displayed for comparison in black (N = 212). d Trial-wise early learning. Each line depicts performance changes (micro-offline in red, micro-online in blue, total in black) per trial (mean + s.e.m.). Even under unpredictable-practice period duration, total learning was closely accounted for by micro-offline gains (black and red lines) whereas micro-online performance changes fluctuate around 0 (blue line). c Data points in the violin plot depict the sum of changes in performance over early learning trials in each participant, the red line denotes the mean. In both groups (10 and 5 s), total early learning was accounted for by performance improvements during rest periods, but not during practice periods. ***P < 0.001, two-tailed, one-sample (within group) nonparametric permutation test for each learning partition. No across group comparison was significant. Note that participants with a high performance were selected in both groups due to required two correct sequences in each trial for calculation of microscale learning (“Methods”, “Data Analysis” section), thus trial 1 performance is comparably higher than in Experiment 1 and 2.