Matching Learned Causal Effects of Neural Networks with Domain Priors
This work addresses the issue of ensuring neural networks reflect true causal knowledge for expert users in domains like fairness, but it is incremental as it builds on existing causal interpretation methods.
The paper tackles the problem of neural networks learning spurious correlations instead of true causal relationships from training data, and proposes a regularization method to align learned causal effects with domain priors, achieving maintained accuracy and improved robustness on noisy inputs across twelve benchmark datasets.
A trained neural network can be interpreted as a structural causal model (SCM) that provides the effect of changing input variables on the model's output. However, if training data contains both causal and correlational relationships, a model that optimizes prediction accuracy may not necessarily learn the true causal relationships between input and output variables. On the other hand, expert users often have prior knowledge of the causal relationship between certain input variables and output from domain knowledge. Therefore, we propose a regularization method that aligns the learned causal effects of a neural network with domain priors, including both direct and total causal effects. We show that this approach can generalize to different kinds of domain priors, including monotonicity of causal effect of an input variable on output or zero causal effect of a variable on output for purposes of fairness. Our experiments on twelve benchmark datasets show its utility in regularizing a neural network model to maintain desired causal effects, without compromising on accuracy. Importantly, we also show that a model thus trained is robust and gets improved accuracy on noisy inputs.