method
active
method:logarithm-transformationlogarithm transformation
Parameter-free loss transformation applied to each task loss to equalize scales
Neighborhood — ranked by edge-count
Frameworks (1)
framework
- IMTL-Lassociated_withPrior loss-balancing method using learnable loss transformation; logarithm approach recovers this
Concepts (1)
concept
- loss-scale balancingassociated_withimplementsAddressing disparity in loss magnitudes across tasks at the loss level
Methods (1)
method
- The proposed method combining loss-scale balancing via logarithm transformation and gradient-magnitude balancing via maximum-norm normalization.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Compared to IMTL-L: parameter-free, no extra computational cost, achieves same theoretical goal
- The logarithm transformation is simpler and more effective than IMTL-L because it is parameter-free.claim0.774Comparison of loss-scale balancing techniques.
- Generalization of the loss transformation.
- A transformation that sharpens and increases the distinction between two types of centers, creating stronger polarity.
- We find that the logarithm transformation also benefits existing gradient balancing methods.quote0.749Key finding showing the broader utility of the log transformation.
- A domain-specific language for linear transformations, specified by denotation as linear maps (a ⊸ b).
- A transformation that develops a thick boundary zone around a zone to intensify its coherence.