Empirical sweeps map scaling of in-context classification accuracy and benign overfitting in transformers on synthetic Gaussian-mixture tasks as functions of dimension, context size, and task diversity.
What Can Trans- formers Learn In-Context? A Case Study of Simple Function Classes
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Investigation into In-Context Learning Capabilities of Transformers
Empirical sweeps map scaling of in-context classification accuracy and benign overfitting in transformers on synthetic Gaussian-mixture tasks as functions of dimension, context size, and task diversity.