AWhat changed
Extended straight-through low-bit network training to separately configurable weight/activation/gradient precision, enabling bit-convolution computations in forward and backward passes.
Scientific scope
Extended straight-through low-bit network training to separately configurable weight/activation/gradient precision, enabling bit-convolution computations in forward and backward passes. Inherits BinaryNet/XNOR and STE; retained higher-precision optimization state and special treatment of layers must not be described as entirely binary training.1/2/4-bit example is a tested configuration, not a universal accuracy-preserving setting; hardware potential is not an end-to-end speedup on every device. Source Eq. 1's bitcount shorthand omits the signed-dot-product affine correction; no new exact signed identity certified.
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 BinaryNet/XNOR and STE; retained higher-precision optimization state and special treatment of layers must not be described as entirely binary training.1/2/4-bit example is a tested configuration, not a universal accuracy-preserving setting; hardware potential is not an end-to-end speedup on every device. Source Eq. 1's bitcount shorthand omits the signed-dot-product affine correction; no new exact signed identity certified.
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