Neural Additive Models: Interpretable Machine Learning with Neural Nets
This addresses the need for interpretable machine learning in high-stakes domains like healthcare, offering a novel hybrid approach that balances accuracy and transparency.
The paper tackles the problem of interpretability in deep neural networks by proposing Neural Additive Models (NAMs), which combine the expressivity of DNNs with the intelligibility of generalized additive models, achieving accuracy similar to state-of-the-art interpretable models and demonstrating flexibility in multitask learning on datasets like COMPAS and COVID-19.
Deep neural networks (DNNs) are powerful black-box predictors that have achieved impressive performance on a wide variety of tasks. However, their accuracy comes at the cost of intelligibility: it is usually unclear how they make their decisions. This hinders their applicability to high stakes decision-making domains such as healthcare. We propose Neural Additive Models (NAMs) which combine some of the expressivity of DNNs with the inherent intelligibility of generalized additive models. NAMs learn a linear combination of neural networks that each attend to a single input feature. These networks are trained jointly and can learn arbitrarily complex relationships between their input feature and the output. Our experiments on regression and classification datasets show that NAMs are more accurate than widely used intelligible models such as logistic regression and shallow decision trees. They perform similarly to existing state-of-the-art generalized additive models in accuracy, but are more flexible because they are based on neural nets instead of boosted trees. To demonstrate this, we show how NAMs can be used for multitask learning on synthetic data and on the COMPAS recidivism data due to their composability, and demonstrate that the differentiability of NAMs allows them to train more complex interpretable models for COVID-19.