Interactive Development of Adaptive Learning and Memory

Abstract

Across development, interactions between value-based learning and memory processes promote the formation of mental models that enable flexible goal pursuit. Value cues in the environment signal information that may be useful to prioritize in memory; these prioritized memories in turn form the foundation of structured knowledge representations that guide subsequent learning. Critically, neural and cognitive component processes of learning and memory undergo marked shifts from infancy to adulthood, leading to developmental change in the construction of mental models and how they are used to guide goal-directed behavior. This review explores how changes in reciprocal interactions between value-based learning and memory influence adaptive behavior across development and highlights avenues for future research.

Pruning recurrent neural networks replicates adolescent changes in working memory and reinforcement learning

“…During the period over which synapses are pruned in cortical areas that support cognition, there is also improvement in several behavioral processes ( 25 ). Two processes that improve are working memory ( 40 ) and RL ( 22 , 27 – 29 , 32 ).…”

“…Cognition and learning also change during adolescence ( 22 – 26 ). For example, there are consistent and systematic changes in performance on reinforcement learning (RL) tasks, in which participants have to learn from feedback to choose rewarding options and avoid punishing options ( 27 – 30 ).…”

Explicitly predicting outcomes enhances learning of expectancy-violating information

“…Expectancy-violating outcomes are assumed to particularly benefit from this enhanced attention because they cannot be easily integrated into existing knowledge structures (Chinn & Brewer, 1993 ). Because this integration process is heavily dependent on top-down attentional control mechanisms mediated by the prefrontal cortex (Brod, Lindenberger, & Shing, 2017 ; Hartley, Nussenbaum, & Cohen, 2021 ), it is further unclear whether the memory benefit is subject to developmental changes.…”

Reappraising the role of dopamine in adolescent risk-taking behavior

“…Importantly, both the interpretations of the meaning of dopamine signals as a general prediction error and as a precision signal in the active inference framework can expand the proposed role of dopamine in the development of mental models (Collins and Cockburn, 2020;Daw, 2018;Sharpe et al, 2017;Daw et al, 2011). Moreover, they also fit the emerging views that emphasize the interplay between learning and memory in improving decisionmaking during adolescence and that propose a role for dopamine in this process beyond a simple relationship with rewards (Murty et al, 2016;Romer et al, 2017;Hartley et al, 2021).…”

“…Uncertainty may play an even larger role considering theories linking dopamine not only to the learning of simple associations between stimulus/behaviors and rewards, but also to forming mental models to guide future actions (Daw et al, 2011;Sharpe et al, 2017;Daw, 2018;Collins and Cockburn, 2020). This makes ecological sense, as highlighted in recent developmental models (Murty et al, 2016;Romer et al, 2017;Hartley et al, 2021): Adolescents lack life-experience but are rapidly expected to assume a myriad of new social roles and responsibilities, so that this is an age that requires intense capacity to learn how to behave based on each new decision-making experience. This view of dopamine as a learning signal and a marker of probabilistic representations can help to explain behavioral evidence which suggests that adolescents make more choices associated with possible negative outcomes compared with adults, but only in situations that involve uncertainty about what might happen (Tymula et al, 2012;Romer et al, 2017).…”