Search papers, labs, and topics across Lattice.
This paper addresses the issue of omitted variable bias in deep learning models, particularly in the context of shortcut learning, where demographic variables inadvertently influence predictions. The authors propose a novel control variable approach that integrates generalized additive modeling with a cross-fitting estimation procedure to refit the final layer of pre-trained networks, effectively orthogonalizing covariate effects. Their method not only yields unbiased predictions but also demonstrates superior performance in recovering true effects compared to existing methods, validated through both simulated and real neuroimaging data.
Omitted variable bias in deep learning can be effectively controlled, leading to unbiased predictions and improved model performance even in the presence of confounding variables.
Control variables are widely used in statistical modelling to account for omitted variable bias of known confounders. However, they have largely been underexplored in deep learning. This is surprising, given that deep learning models encode image-inferable covariates, such as demographic variables, into their predictions when these covariates are correlated with the outcome---a form of omitted variable bias referred to as'shortcut learning'. While many existing confound-control or fairness methods try to restrict the correlation of such covariates with model predictions, we show that this fails to correct for omitted variable bias. We therefore propose a control variable approach for deep learning models, based on generalised additive modelling of the effects of model inputs and covariates. As flexible additive models can suffer from concurvity, we introduce an estimation procedure that refits the final layer of a pre-trained network to include covariate effects, using cross-fitting with ridge penalisation. We show how these effects can be orthogonalised with respect to covariates to exclude their mediated effects and that model predictions can be marginalised over the covariate distribution to control for their effect. This yields unbiased, interpretable predictions and offers flexibility to model the desired effects depending on the scientific or fairness objective. We verify our approach using simulated images, and demonstrate consistent estimation of true effects. Existing methods either require more data or fail to recover the true effects. We apply our method to real neuroimaging data with experimentally induced confounding, where it recovers prediction performance to near the level of a model trained on unconfounded data. Code is available at https://github.com/mpff/cocodeel.
