Disentangling deep neural networks with rectified linear units using duality

TL;DR

This paper introduces deep linearly gated networks (DLGN) as an interpretable alternative to traditional deep neural networks with ReLUs, disentangling computations into interpretable linear components and analyzing their properties.

Contribution

The paper proposes DLGN, a novel interpretable model that disentangles neural network computations into linear components, and analyzes the neural path kernel's invariance properties.

Findings

DLGN recovers over 83.5% of DNN performance on CIFAR datasets.

Convolution with global pooling provides rotational invariance to the neural path kernel.

Skip connections induce ensemble-like structure in the neural path kernel.

Abstract

Despite their success deep neural networks (DNNs) are still largely considered as black boxes. The main issue is that the linear and non-linear operations are entangled in every layer, making it hard to interpret the hidden layer outputs. In this paper, we look at DNNs with rectified linear units (ReLUs), and focus on the gating property (`on/off' states) of the ReLUs. We extend the recently developed dual view in which the computation is broken path-wise to show that learning in the gates is more crucial, and learning the weights given the gates is characterised analytically via the so called neural path kernel (NPK) which depends on inputs and gates. In this paper, we present novel results to show that convolution with global pooling and skip connection provide respectively rotational invariance and ensemble structure to the NPK. To address `black box'-ness, we propose a novel interpretable counterpart of DNNs with ReLUs namely deep linearly gated networks (DLGN): the pre-activations to the gates are generated by a deep linear network, and the gates are then applied as external masks to learn the weights in a different network. The DLGN is not an alternative architecture per se, but a disentanglement and an interpretable re-arrangement of the computations in a DNN with ReLUs. The DLGN disentangles the computations into two `mathematically' interpretable linearities (i) the `primal' linearity between the input and the pre-activations in the gating network and (ii) the `dual' linearity in the path space in the weights network characterised by the NPK. We compare the performance of DNN, DGN and DLGN on CIFAR-10 and CIFAR-100 to show that, the DLGN recovers more than $83.5\%$ of the performance of state-of-the-art DNNs. This brings us to an interesting question: `Is DLGN a universal spectral approximator?'