Contributions of statistical learning to learning from reward feedback

Abstract

AbstractNatural environments are abundant with patterns and regularities. These regularities can be captured through statistical learning, which strongly influences perception, memory, and other cognitive functions. By combining a sequence-prediction task with an orthogonal multidimensional reward learning task, we tested whether detecting environmental regularities can also enhance reward learning. We found that participants used regularities about features from the sequence-prediction task to bias their behavior in the learning task. Fitting choice behavior with computational models revealed that this effect was more consistent with attentional modulations of learning, rather than decision making. Specifically, the learning rates for the feature with regularity were higher, particularly when learning from forgone options during unrewarded trials, demonstrating that statistical learning can intensify confirmation bias in reward learning. Overall, our findings suggest that by enhancing learning about certain features, detecting regularities in the environment can reduce dimensionality and thus mitigate the curse of dimensionality in reward learning.Significance statementNatural environments are filled with detectable patterns and regularities, which, once identified through statistical learning, engage our attentional system and significantly influence multiple cognitive functions. This study explores whether these processes can enhance reward learning in high-dimensional environments with limited reward feedback. Using a novel experimental paradigm and computational methods, we discovered that detecting regularities in specific stimulus features increases learning rates for those features, especially for unrewarded, forgone options. Our findings suggest that identifying environmental regularities can improve feature-based learning and mitigate the curse of dimensionality.