A phased, human-guided multi-fidelity HPO framework with threshold meta-learning was applied to JEREX variants; Longformer with dynamic task loss weighting achieved the best relation F1 of about 49.6 on Re-DocRED.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Interim Report on Human-Guided Adaptive Hyperparameter Optimization with Multi-Fidelity Sprints
A phased, human-guided multi-fidelity HPO framework with threshold meta-learning was applied to JEREX variants; Longformer with dynamic task loss weighting achieved the best relation F1 of about 49.6 on Re-DocRED.