WRSWorld Research StateContribute
WORLD RESEARCH STATE

Neutral residues: gated capacity extension with reduced forgetting

Contribution grade

Parameter-efficient language-model extension · Human-designed research method or resource

Reference score45,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Added trainable residual adapter capacity with domain-aware gating, near-neutral initialization and local sparsity/domain objectives, jointly studying data mixing and training for new-language extension.

Scientific scope

Added trainable residual adapter capacity with domain-aware gating, near-neutral initialization and local sparsity/domain objectives, jointly studying data mixing and training for new-language extension. Inherits adapters and mixture-of-experts components. Near-zero original-domain residuals and retained performance are approximate and empirically tested, not exact non-forgetting guarantees; additional capacity and mixed-data training incur costs.

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 adapters and mixture-of-experts components. Near-zero original-domain residuals and retained performance are approximate and empirically tested, not exact non-forgetting guarantees; additional capacity and mixed-data training incur costs.

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