Neural Additive Models: Interpretable Machine Learning with Neural Nets

By Rishabh Agarwal, Nicholas Frosst, Xuezhou Zhang, R. Caruana, Geoffrey E. Hinton. 2020

612 citations recorded by the source corpus.

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 can be more easily applied to real-world problems.

Subfields: Artificial Intelligence, Model Architectures, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Model Architectures

In Safety, Robustness & Interpretability

Citations in the corpus

Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges (2021, methodology): “Neural networks like neural additive models [5] and neural decision trees [322] could potentially be used to model c → y, since they are both differentiable, nonlinear, and intrinsically interpretable once the input features are interpretable.”
Partially Interpretable Estimators (PIE): Black-Box-Refined Interpretable Machine Learning (2021, methodology): “, 2013), or neural networks in a recent work Neural Additive Models (NAM) (Agarwal et al., 2020).”
Square One Bias in NLP: Towards a Multi-Dimensional Exploration of the Research Manifold (2022, background): “As a result, researchers eschew directions focusing on models that are intrinsically more interpretable such as generalized additive models (Hastie and Tibshirani, 2017) and their extensions (Chang et al., 2021; Agarwal et al., 2021) but which have so far not been shown to match the performance of s”
Why do tree-based models still outperform deep learning on typical tabular data? (2022, positive): “Note also that our observation could also explain the benefits of the ExU activation used in the Neural-GAM paper [Agarwal et al., 2021], and of the embeddings used in Gorishniy et al.”
Benchmarking and survey of explanation methods for black box models (2021, background): “NAM, Neural Additive Models (Agarwal et al. 2021) is a local model-specific explainer defined as a variant of gam but tailored for neural networks. nam aims at combining the performance of deep neural networks with the inherent intelligibility of gam. nam is able to learn graphs that describe how th”
An interpretable bimodal neural network characterizes the sequence and preexisting chromatin predictors of induced transcription factor binding (2021, background): “concept of a Neural Additive Model [64], which linearly combines neural networks that each attend to a single input feature or feature type.”
SurvNAM: The machine learning survival model explanation (2021, methodology): “However, the most interesting method based on GAM is the Neural Additive Model (NAM) [14].”
Trust in AI: progress, challenges, and future directions (2024, neutral): “…of transparency and explainability, meaning that models with the clearest explanations, such as decision trees, may not have a good performance, while those that are the most accurate, such as deep learning based models, are the least explainable (Agarwal et al., 2021; Holzinger et al., 2017).”
Obtaining genetics insights from deep learning via explainable artificial intelligence (2022, background): “Recently Agarwal and colleagues 82 introduced a simplified neural network architecture called neural additive models (NAMs), which sacrifice some of the capacity of traditional neural network models to gain direct parameter interpretation.”
KAN: Kolmogorov-Arnold Networks (2024, positive): “The understanding of these mobility edges is crucial for explaining various fundamental phenomena such as the metal-insulator transition in solids [35], as well as localization effects of light in photonic devices [36, 37, 38, 39, 40].”
Opportunities and Challenges in Explainable Artificial Intelligence (XAI): A Survey (2020, background): “2007 Neural Additive Model, 2020, [50] GL Other IN”
When Do Neural Nets Outperform Boosted Trees on Tabular Data? (2023, result): “We also include partial results for NAM [1], DeepFM [28], and TabTransformer [34].”
Why do tree-based models still outperform deep learning on tabular data? (2022, neutral): “Note also that our observation could also explain the beneﬁts of the ExU activation used in the Neural-GAM paper [Agarwal et al., 2021], and of the embeddings used in Gorishniy et al. [2022]: the periodic embedding might help the model to learn the high-frequency part of the target function, and…”
Interpretable deep learning: interpretation, interpretability, trustworthiness, and beyond (2021, methodology): “Self-interpretable models Capsule [73, 142], Neural additive models [5], CALM [88]”