Devika Narain

Humans can generalize successfully and rapidly after observing just a few training examples. Recent work has attributed this kind of inductive learning to the inference of structure among variable relationships in the world. This means that humans can isolate low dimensional maps that guide and constrain their search in higher-dimensional spaces. This framework is referred to as structure learning. We study how humans abstract information about the structure of variables in the world and how they infer underlying models. We develop generative algorithms for how model hypothesis are extracted by humans from noisy observations and what computational mechanism underlie their inference behavior