I came across the concept of 'transfer learning' in artificial intelligence today. This is not 'knowledge transfer' from one person to another. Rather, the idea is to teach an AI to apply knowledge it learning in one domain to speed its learning in another domain. This article is a deep deep dive into the concept - my advice is to start reading and stop when you feel like stopping. You'll get the gist; the article is pretty well written. I found it interesting because there are theorists (I'm thinking especially of Willingham) who argue for the primacy of 'content knowledge' because you can't apply principles from one domain to another - principles like, say, mathematics or critical thinking. But if an artificial neural network can figure it out, odds are, so can a human. More on transfer learning: this key paper on pretrained language models, this paper on tweet stance classification, and this paper on Google's BERT, a transfer learning model.
Today: 0 Total: 17 [Direct link] [Share]
