A multitask influence function that estimates per-sample cross-task influence is derived and shown to approximate leave-one-out retraining, enabling data pruning that slightly improves multitask accuracy.
Title resolution pending
1 Pith paper cite this work, alongside 19 external citations. Polarity classification is still indexing.
1
Pith paper citing it
19
external citations · OpenAlex
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Measuring Fine-Grained Relatedness in Multitask Learning via Data Attribution
A multitask influence function that estimates per-sample cross-task influence is derived and shown to approximate leave-one-out retraining, enabling data pruning that slightly improves multitask accuracy.