a prelude

This class serves as a foundational introduction to neural networks, with two research foci:

• theoretical frameworks, Omar
• computer vision. Meri

The class proceeds with a self-paced strategy. This means, you get the creative freedom to work on what you find both interesting and challenging! That being said, you will be working independently during class. We will be there in strategic moments to get you moving forward.

Based on your progress and interests, you will be assigned a project. The final deliverable is an article plus a presentation. All proceedings go to the Muffin seminar!

> the muffin seminar 🧁

projects

Below is a diverse set of projects. Each project contains a precise set of deliverables, accompanied with useful resources.

> the vanishing gradient problem

This project aims to explore why vanishing gradients occur in deep neural networks, how they impact training, and strategies to mitigate them using calculus and AI concepts.

> neural ordinary differential equations

The goal is to explain how Neural ODEs unify calculus and AI by replacing traditional neural network layers with continuous dynamics modeled by ordinary differential equations.

> big-\(\mathcal{O}\) analysis of convolutional neural network architecture

Here, we study how convolutional neural networks process images, and how to reason about their computational efficiency using Big−O notation.

> topological sort and neural network pruning

One studies topological sorting, an algorithmic method from graph theory, to prune neural networks efficiently. This is done by understanding dependencies between layers or neurons.

gallery

Final day of class, many finalizing their projects… and some just getting started… (I see you Shayan)

Afternoon classes are for assignments and independent work. At least in principle…

The up-to-down learning approach in action.