Introduction to Deep Learning

What you will be able to do

• Derive linear and logistic regression and generalized linear models from maximum likelihood.
• Explain how backpropagation computes gradients, and implement it by hand.
• Build and train feed-forward networks and CNNs for images and time series, in Keras and JAX.
• Diagnose why a network trains badly or overfits.

Prerequisites

Experience in software development (not necessarily Python) and a basic understanding of statistics, linear algebra and calculus (BSc in exact sciences or similar).

Outline

Duration: 3 days, on-site or remote. Instruction with code, built-in exercises and laptops open with the source code throughout. Work is done in Python, in interactive notebooks.

1. Maximum likelihood estimation
2. Linear models and generalized linear models: linear, logistic and softmax regression
3. Feed-forward networks and backpropagation in JAX, built up step by step
4. Feed-forward networks in Keras, and diagnosing under- and overfitting
5. Convolutional neural networks, for images
6. Convolutional neural networks, for time series

Sample materials

• Maximum likelihood estimation
• Linear regression
• Logistic regression
• Softmax model
• Feed forward networks
• Feed forward networks with Keras
• Convolutional networks with Keras
• Convolutional neural networks for time series

More in yoavram/DataSciPy.

Related workshops

• Advanced Deep Learning
• Open-Set Learning
• LLMs from First Principles