We find that the logarithm transformation also benefits existing gradient balancing methods.

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

• The logarithm transformation also benefits existing gradient balancing methods.claim0.964
  Generalization of the loss transformation.
• Logarithm transformation is simpler and more effective than learnable loss transformationclaim0.822
  Compared to IMTL-L: parameter-free, no extra computational cost, achieves same theoretical goal
• The proposed gradient-magnitude balancing method consistently outperforms GradNorm, as it guarantees equal gradient magnitudes and considers update magnitude.claim0.814
  Advantage over GradNorm.
• The logarithm transformation is simpler and more effective than IMTL-L because it is parameter-free.claim0.810
  Comparison of loss-scale balancing techniques.
• The logarithm transformation (loss-scale balancing) consistently outperforms IMTL-L on NYUv2, Cityscapes, Office-31, Office-Home.finding0.808
  Comparison of loss-scale balancing with IMTL-L.
• DB-MTL achieves loss-scale balancing by performing logarithm transformation on each task loss, and rescales gradient magnitudes by normalizing all task gradients to comparable magnitudes using the maximum gradient norm.quote0.801
  Concise summary of the DB-MTL method from the abstract.
• Logarithm transformation improves PCGrad, GradVac, IMTL-G, CAGrad, Nash-MTL, and Aligned-MTL on NYUv2 (Figure 1).finding0.795
  Effectiveness of logarithm transformation as a plug-in for gradient balancing methods.
• Loss-scale balancing and gradient-magnitude balancing are complementary and combining them achieves the best performance.claim0.787
  Ablation conclusion.