Creating Powerful and Interpretable Models with Regression Networks

Abstract

As the discipline has evolved, research in machine learning has been focused more and more on creating more powerful neural networks, without regard for the interpretability of these networks. Such "black-box models" yield state-of-the-art results, but we cannot understand why they make a particular decision or prediction. Sometimes this is acceptable, but often it is not. We propose a novel architecture, Regression Networks, which combines the power of neural networks with the understandability of regression analysis. While some methods for combining these exist in the literature, our architecture generalizes these approaches by taking interactions into account, offering the power of a dense neural network without forsaking interpretability. We demonstrate that the models exceed the state-of-the-art performance of interpretable models on several benchmark datasets, matching the power of a dense neural network. Finally, we discuss how these techniques can be generalized to other neural architectures, such as convolutional and recurrent neural networks.

A Neural Phillips Curve and a Deep Output Gap

“…HNN-F could be seen to be on the fringe of it, with its multiplicative effects that would certainly be an odd modeling choice without a time-varying unobserved components regression in mind. Closely related, Agarwal et al (2020), O'Neill et al (2021), and Rügamer et al (2020 all develop an architecture inspired from generalized additive models to enhance interpretability in deep networks for generic tasks. While these articles certainly tackle some of the opacity issues coming from nonparametric nonlinear estimation with deep learning, none address those that are inherent to any non-sparse high-dimensional (even linear) regression-i.e., that analyzing partial derivatives of 200 things that typically co-move together unfortunately borders on the meaningless.…”

Attention-like feature explanation for tabular data

“…Similar approaches using neural networks to construct GAMs and to perform shape functions are the basis of methods called GAMI-Net [16] and AxNNs [17]. An architecture called the regression network which can be also regarded as a modification of NAM is proposed by O'Neill et al [61].…”

“…It is important to note that the number of neural networks to take into account pairwise interactions does not increase in the proposed approach. This is a very important difference of the proposed approach from NAM [15] and its modifications, for example, regression networks [61]. Another advantage of the above approach in comparison with other approaches is that we get shape functions of pairs of features instead of some coefficients.…”

“…Another important problem, which can be met in explanation, is a lack of efficient methods for identifying interactive effects or interactions between features when only combinations of two and larger features explain predictions of a black-box model. This problem has been partially solved in NAM and its modifications [15,61,16] where pairs of features can be fed to separate subnetworks and provide shape functions of two variables as outputs of these networks. The main obstacle for implementing NAM with pairs of features is that a huge number of subnetworks should be trained.…”