Groups
Category
In-context learning (ICL) means a model learns from examples provided in the input itself, without updating its parameters.
Statistical learning theory explains why a model that fits training data can still predict well on unseen data by relating true risk to empirical risk plus a complexity term.