AWhat changed
Replaced encoder self-attention with real-valued outputs of Fourier transforms over sequence and hidden dimensions, retaining feed-forward blocks and studying speed–accuracy tradeoffs.
Scientific scope
Replaced encoder self-attention with real-valued outputs of Fourier transforms over sequence and hidden dimensions, retaining feed-forward blocks and studying speed–accuracy tradeoffs. Fourier mixing is not an approximation that preserves each attention matrix or output. FFT and dense-DFT implementations have different complexity and device tradeoffs. Reported speedups use particular encoder tasks and hardware and do not establish drop-in causal decoding equivalence.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Fourier mixing is not an approximation that preserves each attention matrix or output. FFT and dense-DFT implementations have different complexity and device tradeoffs. Reported speedups use particular encoder tasks and hardware and do not establish drop-in causal decoding equivalence.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction