While on paternity leave, I’m boning up on machine learning as a side interest. Watching babies learn at the same time provides some really interesting comparisons.

I’m particularly struck by how babies learn new physical skills like standing and walking, and all of the failed attempts they go through. This seems somehow similar to the process of training neural nets. In contrast, you can also see them learn new categories of things from a single example (e.g. food I like). The speed with which they generalize is astounding!