Lectures, Deep learning course, 2024

Notes:

= Interactive visualization

= static visualization

= video

= book

= paper

= article

Preface and motivation (Lecture 1)

• Deep Learning textbook by Ian Goodfellow, Yoshua Bengio and Aaron Courville: Chapter 1 - Introduction
• Understanding Deep Learning book by Simon J.D. Prince: Chapter 1. The book is currently (checked on 23/01/2024) available for a free download (download page)
• History of Netflix Prize (Wikipedia)
• History of ImageNet competition (Wikipedia), AlexNet paper
• History of neural networks (link)
• FP8 (floating-point 8-bit) number format (announcement) and technical (description)

Logistic regression (Lecture 2)

Definition

• Logistic Regression in Python by RealPython
• Coding interpretation of entropy (Wikipedia)
• Understanding Binary Cross-Entropy / Log Loss in 5 minutes (YouTube)

Learning by Gradient Descent (GD)

• GD algorithm by RealPython
• GD visualization at deeplearning.ai

Singe- and multi-layer neural network

Overview

• A Quick Introduction to Vanilla Neural Networks (link)
• On-line visualization of neural networks at TensorFlow Playground with some explanations from Google
• Hidden representation (learned features) visualization (link)
• Understanding Deep Learning book by Simon J.D. Prince: Chapters 3 and 4 (download page)
• DNN chapter in Hebrew
• But what is a neural network? 3Blue1Brown video

Activation functions

• Comprehensive synthesis of the main activation functions pros and cons (link)
• Commonly used activation functions on stackexchange

Backpropagation and initialization

• Visualization of backpropagation and (Xavier) initialization by Katanforoosh & Kunin, “Initializing neural networks”, deeplearning.ai, 2019
• GD modifications and stochastic GD by Katanforoosh, Kunin et al., “Parameter optimization in neural networks”, deeplearning.ai, 2019

Regularization (Dropout, batch normalization)

• Ridge (L2) Regression (StatQuest)
• Paperswithcode review
• StatQuest video
• Where should place Dropout, Batch Normalization, and Activation Layer?
• Intro to Optimization in Deep Learning: Busting the Myth About Batch Normalization (link)
• Hands-On Batch normalization
• Paper Group Normalization

Scaled Exponential Linear Unit (self-normalizing networks)

• Theory (paperswithcode)
• Note: requires special AlphaDropout and LecunNormal initializer.

Training

GD for NN

• Gradient descent, how neural networks learn 3Blue1Brown video
• Bigger batches are not nessesary better by Keshar at el, On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima, ArXiv, 2017.

CNN and Conv2D

• CNN Explainer online visualization
• 2 Convolution arithmetic
• Understanding 1D, 2D and 3D Convolution Network on Kaggle
• Intuitive understanding of 1D, 2D, and 3D convolutions in convolutional neural networks report
• Hyper-parameters calculation: Convolutional Neural Networks, Explained