AWhat changed
Replaced AdaGrad's unbounded squared-gradient accumulation with a decaying average and used prior-update RMS over gradient RMS to form dimensionally matched coordinate updates.
Scientific scope
Replaced AdaGrad's unbounded squared-gradient accumulation with a decaying average and used prior-update RMS over gradient RMS to form dimensionally matched coordinate updates. Inherits AdaGrad and earlier adaptive-step methods. Original removes a manually set global learning rate but retains decay and numerical-stability hyperparameters; no universal tuning-free or convergence guarantee. The Hessian analogy motivates scaling, not actual exact second-order information or guaranteed optimum.
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 AdaGrad and earlier adaptive-step methods. Original removes a manually set global learning rate but retains decay and numerical-stability hyperparameters; no universal tuning-free or convergence guarantee. The Hessian analogy motivates scaling, not actual exact second-order information or guaranteed optimum.
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