신경 상미분방정식 기반 이산 처치 변수의 평균 용량-반응 함수 추정 기법

초록

When the observable treatment variables are discretely distributed, the absence of support data sample makes estimating average dose-response function further challenging. In this paper, we propose an average dose-response function estimation technique using neural ordinary differential equations. Our experimental result using semi-synthetic dataset empirically demonstrates that utilizing neural ordinary differential equations is about 27% more precise for the discrete treatment scenarios.