Randomizing every treatment independently with probability 1/2 is near-optimal for estimating bounded-degree interaction effects in combinatorial interventions, with an active multi-round rule for small samples.
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
-
Probabilistic Factorial Experimental Design for Combinatorial Interventions
Randomizing every treatment independently with probability 1/2 is near-optimal for estimating bounded-degree interaction effects in combinatorial interventions, with an active multi-round rule for small samples.