Abstract: In recommendation systems, bias is ubiquitous because the data are collected from user behaviors rather than from reasonable experiments. AutoDebias, which resorts to metalearning to find appropriate debiasing configurations, i.e., pseudolabels and confidence weights for all user-item pairs, has been demonstrated to be a generic and effective solution for tackling various biases. Nevertheless, setting pseudolabels and weights for every user-item pair can be a time-consuming process. Therefore, AutoDebias suffers from an enormous computational cost, making it less applicable to real cases. Although stochastic gradient descent with a uniform sampler can be applied to accelerate training, this approach significantly deteriorates model convergence and stability. To overcome this problem, we propose LightAutoDebias (short as LightAD), which equips AutoDebias with a specialized importance sampling strategy. The sampler can adaptively and dynamically draw informative training instances, which results in better convergence and stability than does the standard uniform sampler. Several experiments on three benchmark datasets validate that our LightAD accelerates AutoDebias by several magnitudes while maintaining almost equal accuracy.