Manuscript Psy03-A@1 · Psy03-A.tex · definition · b30eb6e9

Definition: Declared intervention accounting

Intervention budgets and the failure of weight-only dose › A budget vector · lines 532–559

Psy03-A@1#def:budget

For a fixed parameterization, define

B = (D_{W}, D_{b}, D_{M}, D_{mod}, C_{train}),

where

D_{b} = \frac{\norm b ^{'} - b _{1}}{N _{b}}, D_{M} = \frac{1}{N _{W}} i \sum \abs M_{i}^{'} - M_{i},

and

D_{mod} = (\abs a^{'} - a, 1 {φ^{'} \neq = φ}, D_{schedule}) .

The last entry identifies a separately chosen comparison of the training-noise schedules, on an explicitly aligned time or step domain.

Training cost records at least optimizer steps, examples processed, gradient-scoring batches, and the computational implementation. A mask in a dense linear module does not imply sparse execution cost.

Displayed formulas in this passage

unlabeled · line 535
$B = (D_{W}, D_{b}, D_{M}, D_{mod}, C_{train}),$
unlabeled · line 540
$D_{b} = \frac{\norm b ^{'} - b _{1}}{N _{b}}, D_{M} = \frac{1}{N _{W}} i \sum \abs M_{i}^{'} - M_{i},$
unlabeled · line 545
$D_{mod} = (\abs a^{'} - a, 1 {φ^{'} \neq = φ}, D_{schedule}) .$

Nearby in this section

section: A budget vector