A solution to generalized learning from small training sets found in infant repeated visual experiences of individual objects
Researchers have developed a computational model that mimics how infants learn to categorize objects from limited visual data. By analyzing head-camera footage of infants, they observed that object categories are learned through a skewed distribution of experiences, with many images of familiar objects and fewer of novel ones. This 'lumpy' data structure, characterized by high similarity within clusters and variability between them, was found to support generalization to new instances with minimal training, offering insights for both human and machine learning. AI
IMPACT New computational models inspired by infant learning could lead to more efficient AI generalization from small datasets.