• Advestis

Model Pruning in Deep Neural Networks Using the Tensorflow API

P. Desforges. Towards Data Science. February, 11th 2022.

Abstract: One of the most common problems in machine learning is overfitting. This can occur for a variety of reasons. To address this problem, one common solution is to add regularization terms to the model. Another consists in reducing the complexity of the model by reducing the amount of parameters. For an optimal solution, a combination of both approaches should be taken. In this article, we will explore the latter, and more specifically how to incorporate model pruning (which consists in removing superfluous weights in your models) in your Keras Tensorflow models using the Tensorflow Model Optimization API.