3 ms·
I spent a lot of time thinking about this during my PhD. If you think about it analogies are at the heart of kernels and other similarity based methods. You c
by pfisherman 5y ago
I spent a lot of time thinking about this during my PhD. If you think about it analogies are at the heart of kernels and other similarity based methods.
You can totally difference embedding vectors to represent relations. The canonical example from word2vec is basically solving an analogy. The big problem that you run into when applying this stuff more broadly is context, particularly how much context relating the subject and object you want, and which features encode that context. So the problem of abstraction she is talking about maps onto a regularization / feature selection problem.
- SN76477 5y agoAnalogies are a type of cognitive jig. https://blog.cabreraresearch.org/jf https://blog.cabreraresearch.org/jf
- giardini 5y agoOffhand these "jigs" appear to be no more than a particular (yet not fully defined) set of relations.
- yellow_lead 5y agoYou say this factually, but this seems to be the only research group to actually define and use the term.
- imvetri 5y agocontext problem can be solved by letting the AI build analogy about itself, then do transfer learning. Here is the design. A computer, all hardware documentation, all books about computer, this AI ,and an interface between you and the AI.