DiffPrune: Neural Network Pruning with Deterministic Approximate Binary Gates and $L_0$ Regularization

TL;DR

DiffPrune introduces a differentiable approach for neural network pruning using deterministic approximate binary gates and $L_0$ regularization, enabling efficient model sparsity and flexible pruning strategies.

Contribution

It presents a novel deterministic approximation of Bernoulli variables and a method for model selection with binary gates, facilitating differentiable and flexible pruning.

Findings

Effective sparsity achieved with $L_0$ regularization.

Supports arbitrary group sparsity for flexible pruning.

Demonstrated success on neural network pruning tasks.

Abstract

Modern neural network architectures typically have many millions of parameters and can be pruned significantly without substantial loss in effectiveness which demonstrates they are over-parameterized. The contribution of this work is two-fold. The first is a method for approximating a multivariate Bernoulli random variable by means of a deterministic and differentiable transformation of any real-valued multivariate random variable. The second is a method for model selection by element-wise multiplication of parameters with approximate binary gates that may be computed deterministically or stochastically and take on exact zero values. Sparsity is encouraged by the inclusion of a surrogate regularization to the $L_0$ loss. Since the method is differentiable it enables straightforward and efficient learning of model architectures by an empirical risk minimization procedure with stochastic gradient descent and theoretically enables conditional computation during training. The method also supports any arbitrary group sparsity over parameters or activations and therefore offers a framework for unstructured or flexible structured model pruning. To conclude experiments are performed to demonstrate the effectiveness of the proposed approach.