Learning as the unsupervised alignment of conceptual systems

摘要: Concept induction requires the extraction and naming of concepts from noisy perceptual experience. For supervised approaches, as number grows, so does required training examples. Philosophers, psychologists computer scientists have long recognized that children can learn to label objects without being explicitly taught. In a series computational experiments, we highlight how information in environment be used build align conceptual systems. Unlike learning, learning problem becomes easier more systems there are master. The key insight is each concept has unique signature within one system (for example, images) recapitulated other text or audio). As predicted, children’s early form readily aligned By assembling real-word datasets text, images audio, Roads Love propose embedded allows for an unsupervised fashion.