Besides computer vision and image classification, what other use cases/applications are for few-shot learning?
Few-short learning (FSL) can be useful for many (if not all) machine learning problems, including supervised learning (regression and classification) and reinforcement learning.
The paper Generalizing from a Few Examples: A Survey on Few-Shot Learning (2020) provides an overview (including examples of applications and use cases) of FSL. Their definition of FSL provided is based on Tom Mitchell's famous definition of machine learning.
Definition 2.1 (Machine Learning [92, 94]). A computer program is said to learn from experience $E$ with respect to some classes of task $T$ and performance measure $P$ if its performance can improve with $E$ on $T$ measured by $P$.
Here's the definition of FSL.
Definition 2.2. Few-Shot Learning (FSL) is a type of machine learning problems, specified by $E$, $T$ and $P$, where $E$ contains only a limited number of examples with supervised information for the task $T$.
Specific examples of applications of FSL are
- character generation
- drug toxicity discovery
- sentiment classification from short text
- object recognition
An interesting use case is IQ tests, or program synthesis from examples in general.
IQ tests often require you to derive a program from a few examples that can produce a certain output. See for instance https://github.com/fchollet/ARC