Compositional structure appears when wholes reveal patterns their parts cannot

The authors formalize composition through algorithmic information and generate strings and images with prescribed multiscale organization.

Big Tech
Eric Elmoznino · Tejas Kasetty · Blaise Agüera y Arcas · Yoshua Bengio · Guillaume Lajoie

Mila – Quebec AI Institute · Université de Montréal · Google Paradigms of Intelligence Team · LawZero

Research Digest··2 min read
Elmoznino and colleagues define compositionality as structure found in a whole that cannot be recovered from the best models of its parts.

The authors use algorithmic information theory, which describes structure in terms of the shortest program or compressed explanation for data.

Why this paper

From Google Paradigms of Intelligence Team and 3 others

In one line

Compositional structure emerges when parts together reveal shared structure not captured separately.

What we could check

  • ·No code link found
  • ·No weights link found
  • ·No dataset link found
  • ·No compute details found
  • ·No stated limitations found
  • ·No benchmark numbers found

Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.

§

Research Digest

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.

How we workSubscribe