The influence of blocked versus interleaved training regimes and sleep on multi-task learning

The effectiveness of both blocked and interleaved training regimes in human learning has been demonstrated across various tasks and conditions. While recent research suggests that blocked training may support task representation separation and reduce interference in humans, neural networks tend to benefit more from interleaved training. This study investigates whether interleaved training can also support continual learning in humans when consolidation through sleep is possible. Participants learned three tasks under either a blocked or interleaved regime across two experiments: Experiment 1 used non-semantic contextual cues presented before the stimulus, and Experiment 2 used semantically meaningful contextual cues presented after the stimulus. Tests were conducted immediately and after 24 hours in an interleaved format. In Experiment 1, the blocked regime outperformed the interleaved group during learning, but this advantage did not persist in testing. In Experiment 2, the blocked regime maintained higher accuracy across both learning and test phases. No sleep-related improvements in performance were observed for the interleaved groups in either experiment. In Experiment 2, recurrent neural networks trained on participant data did suggest an increase in task separation after sleep, but this was not associated with improvements in task performance, nor was it specific to the interleaved condition. Together, these findings suggest that blocked training or sleep alone do not ensure superior continual learning, which is further determined by the timing and semantic nature of contextual cues.