Guided by the evolutionary developmental perspective, this multi-method, longitudinal study seeks to examine “hyperparameters” that organize young children’s learning. In particular, we adopt computational modeling and a person-centered approach to characterize children’s exploration–exploitation profiles during a reward probability task. We also explored the family socioeconomic precursors and child psychological sequelae of the profiles. Participants were 243 families with a preschooler (Mean age = 4.60 years, 56% girls) from diverse racial and ethnic backgrounds (46% Black, 19% Latinx). Family socioeconomic risks were measured at wave one, and exploration–exploitation and socioemotional functioning were assessed two years later via a reward probability task and teacher report, respectively. Person-centered approach revealed three exploration–exploitation profiles, reflecting (a) a high-random exploration, (b) a high-directed exploration, and (c) a balanced exploration–exploitation profile. Greater family socioeconomic risk was linked to an elevated likelihood of membership in the high-random exploration profile (e.g., Z = −2.37). In turn, the high-random exploration profile exhibited the highest attention problems and greater externalizing problems. Findings have critical implications for understanding the processes through which children adapt to adverse family contexts.