Gradient-based task affinity estimation requires over 40% sample overlap between tasks to recover meaningful structure, below 30% it is indistinguishable from noise.
15 Workshop @ ICLR 2026 Table 10: MTL benefit prediction results
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Information-Theoretic Requirements for Gradient-Based Task Affinity Estimation in Multi-Task Learning
Gradient-based task affinity estimation requires over 40% sample overlap between tasks to recover meaningful structure, below 30% it is indistinguishable from noise.