Continual Visual and Verbal Learning From a Child's Egocentric Input

arXiv 2606.05115·low signal

Jiang, Yang, and Norman train models to learn word meanings from a continuous, temporally structured stream of egocentric (child-perspective) experience, mirroring how children acquire language. The work probes continual multimodal learning without catastrophic forgetting. More cognitive-science than production, but informs grounded continual-learning design.