Certifying Data-Bias Robustness in Linear Regression
This work addresses data-bias robustness for linear models, which is an incremental contribution to improving trust in model outputs.
The paper tackles the problem of certifying whether linear regression models are robust to label bias in training datasets, showing that linear models often exhibit high bias-robustness but also revealing gaps in robustness under certain bias assumptions.
Datasets typically contain inaccuracies due to human error and societal biases, and these inaccuracies can affect the outcomes of models trained on such datasets. We present a technique for certifying whether linear regression models are pointwise-robust to label bias in the training dataset, i.e., whether bounded perturbations to the labels of a training dataset result in models that change the prediction of test points. We show how to solve this problem exactly for individual test points, and provide an approximate but more scalable method that does not require advance knowledge of the test point. We extensively evaluate both techniques and find that linear models -- both regression- and classification-based -- often display high levels of bias-robustness. However, we also unearth gaps in bias-robustness, such as high levels of non-robustness for certain bias assumptions on some datasets. Overall, our approach can serve as a guide for when to trust, or question, a model's output.