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.

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

• DB-MTL is a simple yet effective method that addresses both loss-scale and gradient-magnitude imbalances.claim0.884
  Core claim of the paper.
• DB-MTL reduces gradient conflict and improves task balance compared to EW.claim0.848
  Effect on gradient conflict.
• Loss-scale balancing and gradient-magnitude balancing are complementary and combining them achieves the best performance.claim0.838
  Ablation conclusion.
• DB-MTL has similar per-epoch running time to gradient balancing methods on NYUv2, slower than loss balancing methods.finding0.837
  Computational efficiency comparison.
• The logarithm transformation (loss-scale balancing) consistently outperforms IMTL-L on NYUv2, Cityscapes, Office-31, Office-Home.finding0.833
  Comparison of loss-scale balancing with IMTL-L.
• Task balancing requires simultaneous consideration of both loss scales and gradient magnitudesclaim0.829
  Core interpretive position of DB-MTL: complementarity of loss and gradient perspectives
• DB-MTL increases gradient cosine similarity faster and keeps it positive on Office-31, reducing gradient conflict vs EW.finding0.821
  Analysis of gradient conflict reduction.
• DB-MTL does not affect training stability; losses converge smoothly.claim0.815
  Training stability claim.