WRSWorld Research StateContribute
WORLD RESEARCH STATE

Multiplicative Integration: factorized recurrent information fusion

Contribution grade

Recurrent neural-network architectures · Human-designed research method or resource

Reference score65,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Fused transformed input/state streams by elementwise products with optional additive and bias terms, embedding the design into recurrent cells to alter gating and gradient flow.

Scientific scope

Fused transformed input/state streams by elementwise products with optional additive and bias terms, embedding the design into recurrent cells to alter gating and gradient flow. Inherits second-order/multiplicative RNNs and LSTM/GRU cells. Generalized variant adds small parameter vectors rather than literally zero parameters. Empirical gradient/optimization behavior does not prove vanishing gradients eliminated or universal superiority over additive cells.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits second-order/multiplicative RNNs and LSTM/GRU cells. Generalized variant adds small parameter vectors rather than literally zero parameters. Empirical gradient/optimization behavior does not prove vanishing gradients eliminated or universal superiority over additive cells.

Contributors

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