Conceptdynamic low-rank approximation
Dynamic Low-rank Approximation
A mechanism proposed by MuEdit that, when the shared null space is insufficient, dynamically constructs an approximate null space through low-rank decomposition, providing more feasible directions for parameter updates.
Timeline (last 90 days)
Nov 13
Dynamic low-rank approximation is used to expand the null space when the common null space is insufficient.
Unverified50%
Nov 13
Whether dynamic low-rank approximation performs consistently on larger-scale models remains an open question left by the study.
Unverified50%