Deep-layered machines have a built-in Occam's razor
2 PM, 14 Sep 2026
Thomas Fink shows how a hierarchical composition naturally favours simple outputs, revealing an intrinsic Occam’s razor in layered systems.
Complex systems map many microscopic descriptions onto few macroscopic outcomes. Surprisingly, these input-output maps favor simple outputs. We prove that, in an exactly solvable deep-layered machine, depth drives the output distribution toward exponential decay in complexity. Thus hierarchical composition alone generates an intrinsic Occam's razor, giving an analytical explanation for simplicity bias in learning.
Dr Thomas Fink is the founding Director of the London Institute and Charge de Recherche in the French CNRS. He studied physics at Caltech, Cambridge and École Normale Supérieure. His work includes statistical physics, combinatorics and the mathematics of evolvable systems.
