CoRe: Learning compact representation with prior-guided initialization

doi:10.1016/j.patcog.2026.113040

CoRe: Learning compact representation with prior-guided initialization

2026-01-09

0

PRE

AI

DOI：10.1016/j.patcog.2026.113040

Original

Original request for help

Save

Abstract

En 中文

• A prior-guided initialization is developed to accurately estimate compact representations and preserve information. • Reveals residual designs as CoRe module special cases, offering insight into residual efficacy. • Each component maps to CoRe method, making CoRe module traceable. • CoRe module adapts flexibly to diverse architectures and tasks. • CoRe can replace components without extra parameters or add with minimal overhead.

CoRe: Learning compact representation with prior-guided initialization

Abstract

Journal

Organization