ConceptLearningWithExemplars

The key lies in the brain's graph structure. In this framework, each digit is linked to a set of visual exemplars. When a new input comes in, the brain looks for the closest match among its stored exemplars. If the match is good enough, the brain treats it as input as that version of the digit. If no match exists, or if the brain initially guesses wrong and is corrected, then a new exemplar is stored, so a single node representing a digit can have any number of exemplars.

• Source: 2025-09-16 Machine Learning vs Human Learning： They’re Not the Same

Importantly, the system doesn't bloat endlessly, the brain employs internal processes that act like housekeeping staff for the graph structure. As with other areas in the graph, they look for common attributes which they can bubble up, and they can merge overlapping exemplars, prune away rarely used ones, and refine the overall representation so it remains efficient.

• Source: 2025-09-16 Machine Learning vs Human Learning： They’re Not the Same