Thought Vectors

The way I understood it, thought vectors map a word to numerous sentences that contain it. In effect, the meaning of a word can be constructed. And so it can be expected that as more sentences are linked to a word, the closer we get from the true meaning of that word [Krumins].

Thought vectors can be applied to image recognition, specifically for describing the contents of an image [Goh]:

screen-shot-2017-01-27-at-2-18-14-pm

screen-shot-2017-01-27-at-3-38-35-pm

 

References:
Thought vectors could revolutionize artificial intelligence, A. Krumins, https://www.extremetech.com/extreme/206521-thought-vectors-could-revolutionize-artificial-intelligence
Decoding the thought vector, G. Goh, http://gabgoh.github.io/ThoughtVectors/

Advertisements