Proceedings of the Annual Meeting of the Cognitive Science Society

Humans are not mere observers, passively receiving the information provided by their environment; they deliberatelyengage with their environment, actively participating in the information acquisition stage to improve their learning per-formance. Despite being a hallmark of human cognition, the computational underpinnings of this active (or self-directed)mode of learning have remained largely unexplored. Drawing on recent advances in machine learning, we present aneural-network model simulating the process of learning how to actively learn. To our knowledge, our work is the firstneural-network model of learning to actively learn. Extensive simulations demonstrate the efficacy of our model, partic-ularly in handling high dimensional domains. Notably, our work serves as the first computational account of the recentexperimental finding by MacDonald and Frank (2016) showing that prior passive learning improves subsequent activelearning. Our work exemplifies how a synergistic interaction between machine learning and cognitive science helps de-velop effective, human-like artificial intelligence.