claim

pending-review

claim:task-balancing-requires-simultaneous-consideration-of-both-loss-scales-and-gradient-magnitudes

Task balancing requires simultaneous consideration of both loss scales and gradient magnitudes

lin-2023-dual-balancing.md

Frontmatter (10 fields)

{
  "doc": "lin-2023-dual-balancing.md",
  "context": "Core interpretive position of DB-MTL: complementarity of loss and gradient perspectives",
  "category": "ai",
  "norm_label": "Task balancing requires simultaneous consideration of both loss scales and gradient magnitudes",
  "graphify_id": "claim_dual_perspective",
  "source_file": "lin-2023-dual-balancing.md",
  "imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/lin-2023-dual-balancing/graph.json",
  "extracted_type": "claim",
  "source_location": "§3",
  "graphify_file_type": "claim"
}

Outgoing (0)

None.

Incoming (3)

Supported by (3)

DB-MTL achieves ∆p = +1.05±0.20 and accuracy 94.09% on Office-31, outperforming baselines(finding)
DB-MTL achieves ∆p = +1.15±0.16 on NYUv2, outperforming all baselines including state-of-the-art(finding)
DB-MTL achieves ∆p = −58.10±3.89 on QM9, best among all MTL methods though below STL baseline(finding)

Mentions (1)

papers-typed
lin-2023-dual-balancing.md