zero-shot learning
Table of Contents
type where the model performs decently without seeing training data
- training data
- usually there is some training data, just not the class of data the model needs to classify like, a set of smile, and a set of dog. The model may be able, after learning the 2 sets, to zero-shot predict “smiling dog”.
https://en.wikipedia.org/wiki/Zero-shot_learning
Backlinks
zero-shot learning enables novel case handling
zero-shot learning make the model works decently on novel cases
As foundation model contains generic human behaviour, when the task complys to generic human behaviours, it can be done instantly, as it were performed by a unskilled mediocre human who knows a lot of information
as foundatoin model usually works with an extensible notion of class, zero-shot learning is almost imperative.