Book Image

Neuro-Symbolic AI

By : Alexiei Dingli, David Farrugia
Book Image

Neuro-Symbolic AI

By: Alexiei Dingli, David Farrugia

Overview of this book

Neuro-symbolic AI offers the potential to create intelligent systems that possess both the reasoning capabilities of symbolic AI along with the learning capabilities of neural networks. This book provides an overview of AI and its inner mechanics, covering both symbolic and neural network approaches. You’ll begin by exploring the decline of symbolic AI and the recent neural network revolution, as well as their limitations. The book then delves into the importance of building trustworthy and transparent AI solutions using explainable AI techniques. As you advance, you’ll explore the emerging field of neuro-symbolic AI, which combines symbolic AI and modern neural networks to improve performance and transparency. You’ll also learn how to get started with neuro-symbolic AI using Python with the help of practical examples. In addition, the book covers the most promising technologies in the field, providing insights into the future of AI. Upon completing this book, you will acquire a profound comprehension of neuro-symbolic AI and its practical implications. Additionally, you will cultivate the essential abilities to conceptualize, design, and execute neuro-symbolic AI solutions.
Table of Contents (12 chapters)

Environment and data setup

The main objective of this chapter is to introduce the different mechanisms and thought processes associated with neuro-symbolic programming. This chapter is not designed as a programming crash course for symbolic or deep learning. For this purpose, we will work with the Red and White Wine Dataset (https://www.kaggle.com/datasets/numberswithkartik/red-white-wine-dataset) – publicly available in Kaggle. This dataset consists of 12 features describing different wine characteristics (such as the density and residual sugar, to name a couple) and a binary label representing whether said wine is a red or white wine. Some characteristics that made this dataset ideal for our use case were the following:

  • It has around 6,000 samples, making it ideal for showing the power of NSAI by varying the size of the training data
  • It does not require much data pre-processing or engineering
  • It is a standard binary classification task, making it more straightforward...