The basics of deep learning

1. Introduction

The goal of the present page is to teach the basics of supervised deep learning, oriented to using PyTorch. While there is a quick review, a previous basic understanding of probability, linear algebra, statistics and mathematical analysis is needed.

2. Useful resources

The below listed learning resources are all FREE.

For seeing a broad overview of methods and what to learn, check the AI expert roadmap.

• Maths
• Linear algebra
• Essence of linear algebra. By 3Blue1Brown.Visual and good for a geometric understanding.
• Linear algebra, full college course. By freeCodeCamp.org.Pragmatic.
• Linear algebra. By Steve Brunton.More oriented to data science.
• Analysis
• Essence of calculus. By 3Blue1Brown.Visual.
• Differential equations. By 3Blue1Brown.
• Probability and statistics
• Statistics fundamentals. By StatQuest.Visual, chilled, full of examples and with just the needed jargon.
• Basics of Probability. By jbstatistics.Short, visual, straight to the point.
• Programming
• Python for beginners. By freeCodeCamp.org.Teaches all, with examples and exercises.
• Julia Tutorials (Basic). By the Julia programming language.One of the languages of the future.
• Data science & machine learning
• Python for Data Science. By freeCodeCamp.org.Extense, from the beginning. Teaches multiple useful libraries.
• MLU explain. By Amazon.Visual webpage, teaches multiple basic concepts.
• Machine learning specialization. By Deeplearning.AI.New version of the famous Machine Learning Stanford course, taught by Andrew NG.

2. Maths review

A quick recap of basic maths concepts important for machine learning models.

2.1. Linear algebra

Vectors

Leaving aside the different interpretations of this concept, a vector starts from an origin (where all coordinates intersect), and ends in a point in space, defined by each value of the different dimensions. These values are represented as numbers in an ordered list.

Concretely, for data science, a vector of features stores the different values of a data point, and the vector of weights stores the different values of the parameters of a model.

Matrices

They are a set of vectors, stored by columns.

Operations

All operations on linear algebra arise from basis vectors, which contain a 1 for a coordinate value, and all 0 for the rest.

Linear combinations

When varying a and b values,