How to Improve Deep Learning Model Robustness by Adding Noise

Last Updated on August 28, 2020

Adding noise to an underconstrained neural network model with a small training dataset can have a regularizing effect and reduce overfitting.

Keras supports the addition of Gaussian noise via a separate layer called the GaussianNoise layer. This layer can be used to add noise to an existing model.

In this tutorial, you will discover how to add noise to deep learning models in Keras in order to reduce overfitting and improve model generalization.

After completing this tutorial, you will know:

• Noise can be added to a neural network model via the GaussianNoise layer.
• The GaussianNoise can be used to add noise to input values or between hidden layers.
• How to add a GaussianNoise layer in order to reduce overfitting in a Multilayer Perceptron model for classification.

Kick-start your project with my new book Better Deep Learning, including step-by-step tutorials and the Python source code files for all examples.

Let’s get started.

• Updated Oct/2019: Updated for Keras 2.3 and TensorFlow 2.0.