loss-scale balancing

Addressing disparity in loss magnitudes across tasks at the loss level

Neighborhood — ranked by edge-count

framework

Dual-Balancing Multi-Task Learning
implements
Novel MTL method combining loss-scale and gradient-magnitude balancing

method

logarithm transformation
associated_withimplements
Parameter-free loss transformation applied to each task loss to equalize scales
DB-MTL (Dual-Balancing Multi-Task Learning)
implements
The proposed method combining loss-scale balancing via logarithm transformation and gradient-magnitude balancing via maximum-norm normalization.

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Loss-scale balancing and gradient-magnitude balancing are complementary and combining them achieves the best performance.claim0.804
Ablation conclusion.
Task balancing requires simultaneous consideration of both loss scales and gradient magnitudesclaim0.782
Core interpretive position of DB-MTL: complementarity of loss and gradient perspectives
Task balancingconcept0.762
The problem of ensuring all tasks in MTL perform well, avoiding dominance by some tasks.
gradient-magnitude balancingconcept0.748
Addressing disparity in gradient magnitudes across tasks at the gradient level
Loss Functionconcept0.748
In machine learning, a function measuring the distance between current and desired output; analogous to stress.
Squared Difference Lossmethod0.739
Loss function used in both experiments: sum of squared differences between predicted and target grid
task balancing problemquestion0.738
Core challenge where disparity in loss and gradient scales among tasks leads to performance compromises
The logarithm transformation (loss-scale balancing) consistently outperforms IMTL-L on NYUv2, Cityscapes, Office-31, Office-Home.finding0.731
Comparison of loss-scale balancing with IMTL-L.