🛰️ Daily AI Frontier
‹ back to 2026-07-30

A Compositional Theory of Causally Masked Transformers

Research LLMs & Foundation Models

Ranking

Overall 76
Content 90
Popularity 44

Observed public metrics from 1 member.

Merged summary

TL;DR - This paper develops an algebraic framework for characterizing what finite-precision, causally masked transformers can compute on arbitrary-length inputs. It connects attention mechanisms and numerical semantics to distinct classes of finite-state memory and expressivity.

  • The framework models attention’s prefix summary as finite internal memory that future queries can access.
  • Independent per-head updates compose across layers, enabling systematic derivation of expressivity bounds.
  • Without positional embeddings, sliding-window, modified soft, combined, and standard floating-point soft attention yield progressively richer memory operations.
  • These cases correspond to definite, R-trivial, locally R-trivial, and aperiodic semigroups; the bounds are tight under an explicit free-wiring assumption.

Sources (1)

A Compositional Theory of Causally Masked Transformers

arXiv cs.FL Franz Nowak, Ryan Cotterell, Reda Boumasmoud 2026-07-29 arXiv:2607.26988
Public signals Semantic Scholar citations 0 · Semantic Scholar influential citations 0
Providers: Hugging Face · N/A OpenAlex · N/A Publisher · N/A Semantic Scholar · Citations 0 · Influential citations 0 X · N/A Fetched 2026-08-24 14:31:43.139195 UTC

TL;DR - This paper develops an algebraic framework for characterizing what finite-precision, causally masked transformers can compute on arbitrary-length inputs. It connects attention mechanisms and numerical semantics to distinct classes of finite-state memory and expressivity.

  • The framework models attention’s prefix summary as finite internal memory that future queries can access.
  • Independent per-head updates compose across layers, enabling systematic derivation of expressivity bounds.
  • Without positional embeddings, sliding-window, modified soft, combined, and standard floating-point soft attention yield progressively richer memory operations.
  • These cases correspond to definite, R-trivial, locally R-trivial, and aperiodic semigroups; the bounds are tight under an explicit free-wiring assumption.
item →