Toddler Headcam Footage Trains AI to Learn Words Like a Child
-
Researchers created an AI model that learned words from a toddler's point of view using footage from a headcam the toddler wore.
-
The model learned by associating words and objects that co-occurred, similar to how children learn language.
-
The model was able to identify some images not in the training data, suggesting an ability to generalize knowledge.
-
The model's success hints that associative learning may be enough for children to acquire vocabulary.
-
The findings suggest an alternative method to train AI using less data that mimics human learning.