For min/max readout, two attention heads beat one head by an exponential resource gap, and for n-bit parity and symmetric Boolean functions, heads times polynomial degree must reach the threshold degree, with matching constructions.
Tighter Bounds on the Expressivity of Transformer Encoders
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
Characterizing neural networks in terms of better-understood formal systems has the potential to yield new insights into the power and limitations of these networks. Doing so for transformers remains an active area of research. Bhattamishra and others have shown that transformer encoders are at least as expressive as a certain kind of counter machine, while Merrill and Sabharwal have shown that fixed-precision transformer encoders recognize only languages in uniform $TC^0$. We connect and strengthen these results by identifying a variant of first-order logic with counting quantifiers that is simultaneously an upper bound for fixed-precision transformer encoders and a lower bound for transformer encoders. This brings us much closer than before to an exact characterization of the languages that transformer encoders recognize.
citation-role summary
citation-polarity summary
fields
cs.LG 1years
2026 1verdicts
ACCEPT 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Attention-based representations for multi-task computation
For min/max readout, two attention heads beat one head by an exponential resource gap, and for n-bit parity and symmetric Boolean functions, heads times polynomial degree must reach the threshold degree, with matching constructions.