The Regularization Cookbook: Explore practical recipes to improve the functionality of your ML models

Methodologies and recipes to regularize nearly any machine learning and deep learning model using cutting-edge technologies such as Stable Diffusion, GPT-3, and Unity

Key Features

• Learn how to diagnose whether regularization is needed for any machine learning model
• Regularize different types of ML models using a broad range of techniques and methods
• Get the best of your models using state of the art Computer Vision and NLP

Book Description

Deploying machine learning solutions is all about getting robust results on new, unseen data. To achieve such results, one way is regularization. Regularization can take many forms and can be used in many ways, and not all methods apply to all cases. This book aims at providing the right tools and methods to handle any case properly, with ready-to-use working codes as well as theoretical explanations whenever possible.

After an introduction to regularization and methods to diagnose when to use it, we will start implementing regularization techniques on linear models such as linear and logistic regression, and tree-based models such as random forest and gradient boosting.

The book will then introduce specific regularization methods based on data. High cardinality features and imbalanced datasets may require specific regularization methods that will be explored.

In the last four chapters, the book will cover regularization for deep learning models. After reviewing general methods that apply to any type of neural network, the book will dive into more NLP-specific methods for RNNs and transformers, as well as using BERT or GPT-3. We will close with regularization for Computer Vision, covering CNN specifics, as well as the use of generative models such as GANs and Stable Diffusion, and third-party software like Unity.

What you will learn

• How to diagnose overfitting properly and when regularization is needed
• Regularizing common linear models such as logistic regression
• Get a deeper knowledge of regularizing tree-based models such as XGBoost
• Leverage structured data to regularize ML models
• Learn general techniques to regularize deep learning models
• Discover specific regularization techniques for NLP problems using Transformers
• Understand the regularization in Computer Vision models and CNN architectures
• Apply cutting-edge computer vision regularization with generative models

Who This Book Is For

Whether you are a data scientist, a machine learning engineer, or just a machine learning enthusiast, if you want to get hands-on knowledge of the available methods to improve the performances of your models, this book is for you.

Basic, hands-on knowledge of Python is expected to get the most out of the proposed codes. Also, basic concepts of ML and DL are reminded to smooth the learning curve, no matter their level. This book is also aimed at experienced professionals willing to use state-of-the-art methods for regularization.

Table of Contents

1. Overview of Regularization
2. Regularization in ML
3. Regularization with ML models
4. Regularization with Tree based models
5. Regularization with structured data
6. Deep learning reminders
7. Regularization of a vanilla neural network
8. Regularization in NLP
9. Regularization in Computer Vision
10. Advanced Regularization in Computer Vision: synthetic image generation

After a Ph.D. in Physics, Vincent Vandenbussche has worked for a decade in the industry, deploying ML solutions at scale. He has worked in numerous companies, such as Renault, L'Oréal, General Electric, Jellysmack, Chanel, and CERN.He also has a passion for teaching: he co-founded a data science bootcamp, was an ML lecturer at Mines Paris engineering school and EDHEC business school and trained numerous professionals in companies like ArcelorMittal and Orange.