Discovering Invariances in Healthcare Neural Networks
arXiv:1911.03295
Abstract
We study the invariance characteristics of pre-trained predictive models by empirically learning transformations on the input that leave the prediction function approximately unchanged. To learn invariant transformations, we minimize the Wasserstein distance between the predictive distribution conditioned on the data instances and the predictive distribution conditioned on the transformed data instances. To avoid finding degenerate or perturbative transformations, we add a similarity regularization to discourage similarity between the data and its transformed values. We theoretically analyze the correctness of the algorithm and the structure of the solutions. Applying the proposed technique to clinical time series data, we discover variables that commonly-used LSTM models do not rely on for their prediction, especially when the LSTM is trained to be adversarially robust. We also analyze the invariances of BioBERT on clinical notes and discover words that it is invariant to.
arXiv commentThe extended version
References in corpus (10)
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- BioBERT: a pre-trained biomedical language representation model for biomedical text mining
- Axiomatic Attribution for Deep Networks
- Interpretable machine learning: definitions, methods, and applications
- On the Convergence of Adam and Beyond
- Sanity Checks for Saliency Maps
- Fast Computation of Wasserstein Barycenters
- Learning with a Wasserstein Loss
- Exploiting Excessive Invariance caused by Norm-Bounded Adversarial Robustness
- Temporal-Clustering Invariance in Irregular Healthcare Time Series