Logarithm transformation is simpler and more effective than learnable loss transformation

{
  "doc": "lin-2023-dual-balancing.md",
  "context": "Compared to IMTL-L: parameter-free, no extra computational cost, achieves same theoretical goal",
  "category": "ai",
  "norm_label": "Logarithm transformation is simpler and more effective than learnable loss transformation",
  "graphify_id": "claim_log_transformation_superiority",
  "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": "§3.1",
  "graphify_file_type": "claim"
}