1 paper
Gérard Biau, Claire Boyer
We propose a statistical proxy framework for retrieval-augmented generation (RAG) that formalizes how language models balance internal predictions with retrieved evidence. We deriv…