124 - Jay McClelland: Deep Learning, Neural Networks, and Artificial Intelligence
Robinson's Podcast
00:00
The Perception of Letters in Words
The model has neuron-like processing units in it for features of letters in each of the possible positions within a word. And so when you see, for example, W-O-R, and you know there's one more letter, but you're not quite sure what it is, a couple of words get activated in your mind. The word word itself and the word work are consistent with those first three letters. If the bottom-up input is maybe consistent with a K or an R, but not with a D, the network will sort of settle into a state where it's decided that this must be work and that final letter must be a K.
Transcript
Play full episode
