Rigorous Theory for General Non‑IID Learning Beyond the Structured Setting

Establish a rigorous theoretical characterization of learning with general non‑i.i.d. data beyond the causal structured k‑gap independent setting analyzed in this work, including precise generalization guarantees for kernel ridge regression.

Background

The authors develop concentration tools and excess risk bounds for KRR under a specific structured non‑i.i.d. model with causal signal–noise structure and k‑gap independence, showing how dependencies and noise multiplicity affect generalization.

They explicitly acknowledge a limitation: their analysis is confined to this structured regime, and they identify the broader task of rigorously characterizing general non‑i.i.d. scenarios as an open problem.