Precision Scaling Laws
Investigating the relationship between network depth, parameter perturbations, and the minimum precision required for acceptable performance.
4 papers
Where this stands
The written synthesis of this thread is for subscribers. Subscribe.
How this thread developed
Oct 2026 · Mutah University
Neural network precision floor follows a depth law predicted by amplification
Derives a depth law for the precision floor of neural networks, relating network depth to the minimum bit-width for acceptable accuracy.
released code
1 further paper
Oct 2026 · Carnegie Mellon University, ML Collective
Compressed looped models settle, not drift, so precise final loops recover them
Reveals that compressed looped models settle to fixed error, enabling recovery with precise final loops.
Oct 2026 · University of Vienna, Huawei Heisenberg Research Center
Adaptive mixed-precision attention matches full precision with selective 16-bit recomputation
Shows that 8-bit attention with selective recomputation matches full precision, informing precision requirements.
3 of 4 papers shown