Session Submission Summary
Share...

Direct link:

3-029 - Understanding infants’ curiosity-based learning: empirical and computational approaches

Sat, May 28, 4:00 to 5:30pm, Hilton New Orleans Riverside, Grand Salon Room 12

Session Type: Paper Symposium

Integrative Statement

As all parents know, infants are curious learners. The vast majority of their time is spent freely exploring a relatively unconstrained learning environment based on their intrinsic curiosity. However, although curiosity-driven exploration accounts for almost all of infants’ experience, our understanding of early development is largely based on tightly-controlled, highly structured experiments. This rich body of empirical and computational work has shown us that learning is exquisitely sensitive to the information structure of the learning environment. However, because these studies are typically conducted under artificially controlled conditions, how infants drive their own learning by sampling their learning environment remains unclear.

This interdisciplinary symposium brings together new insights into infants’ curiosity from developmental psychology, computational modelling and developmental robotics, describing cutting-edge methodologies for both empirical and computational work. Two empirical studies using screen-based eye-tracking in the lab and head-mounted eye-tracking in the wild demonstrate that when allowed to explore freely, infants impose structure on their own learning environment based on systematic information selection. A neurocomputational simulation of infants’ curiosity-based learning offers a new, learner-based mechanism for how curiosity may operate in human infants. Finally, a robotic implementation of curiosity demonstrates autonomous learning in a curiosity-driven system. The synthesis of ideas from multiple disciplines presented in this symposium points to a new approach to understanding infant development, with an essential focus on the infant’s own role in driving learning.

Sub Unit

Chair

Individual Presentations