Prediction and pattern recognition

Abstract

Humans frequently have to learn sequential data through pattern recognition or prediction. Data can come from a wide variety of sources and in a wide variety of forms, such as melodies, numbers or texts. An underlying theory about how humans learn, that is, to make an underlying model of the data generation, is still lacking. In this study, we investigate the usefulness of an information theoretical concept, predictive information decay, in human information processing. A simple experiment is designed and piloted that distinguishes how well people perform on pattern recognition tasks when presented with sequential data with varying predictive information or varying predictive information decay. The underlying data is generated through random walks on k-regular graphs. After tentatively concluding that predictive information or predictive information decay rate do not provide clear correlates to performance on recognition tasks, we provide advice and inside for a more extensive follow-up study.