A radial-basis combiner over base-model outputs, trained with a tunable winner-takes-all diversity weight, is claimed to match loss geometry and beat logit averaging and MoE.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Structured Basis Function Networks: Loss-Centric Multi-Hypothesis Ensembles with Controllable Diversity
A radial-basis combiner over base-model outputs, trained with a tunable winner-takes-all diversity weight, is claimed to match loss geometry and beat logit averaging and MoE.