An alternative account is motivated by findings of ‘‘nonlinear mixed selectivity’’ in PFC (Rigotti et al., 2013). It suggests that PFC is best understood as a random network, in which unique ensembles are activated by different combinations of input features and task contexts (Bouchacourt and Buschman, 2019). This predicts distinct ensembles are activated depending on the route by which information arrives, feedforward versus interhemispheric. Regardless of the specific mechanism, our results indicate that just before the memory trace is to be read out for comparison with a test object, its neural code shifts to become more like the ensemble used for static, feedforward-induced memory traces in the same location. This convergence likely facilitates downstream comparison and decision-making processes by allowing similar mechanisms and readout weights for reading the same information out from WM stores.
