finding
active
finding:log-x-min-s-e-s-x-s-1-for-x-0log(x) = min_s (e^s * x - s - 1) for x > 0
Mathematical equivalence showing logarithm transformation recovers IMTL-L in the limit
Source paper
extracted_from(2023) · Baijiong Lin · Weisen Jiang · Feiyang Ye · Yu Zhang +5
Neighborhood — ranked by edge-count
Claims (1)
claim
- Compared to IMTL-L: parameter-free, no extra computational cost, achieves same theoretical goal
Communities (2)
community
- Dual-balancing multi-task learningmembers_ofDB-MTL jointly balances loss scale and gradient magnitude, benchmarked on NYUv2 and Office-31.
- Parameter-free logarithm transformation for multi-task learning that improves gradient balancing methods like PCGrad and Nash-MTL across vision benchmarks.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- log P(M(α) complies) is approximately linear in α when probability is not close to 1 on Qwen 3 4B pathfinding0.763Empirical basis for the extrapolation.
- Fits log compliance vs α on α with compliance <50% and extrapolates to α=1.
- Core result.
- Logit-minus-mean-logit loss was most effective for attribution, vs log-loss which was confusing to interpretfinding0.737Attribution loss selection.
- Mathematical relationship between IMTL-L and log transformation.
- Justification for the linear combination
- A heuristic measure of degree of life in buildings, combining harmony (H) and temperature (T) to approximate the density of living centers.
- Sampling noise decreases in proportion to 1/√S, giving a linear relation between log TVD and log S.hypothesis0.730Derived from multinomial sampling.