Alexander S. Lundervold, https://alexander.lundervold.com/.

This is the code repository for the deep learning module in the 2022 variant of PCS956 at HVL. Students following the course can access the course Canvas via this link: https://hvl.instructure.com/courses/22784.

This short three-week module will introduce some core features of deep learning, provide hands-on experience with some application areas, and do a guided tour of some valuable tools and techniques of practical deep learning.

The module has three parts:

Part 1 will be an introduction to deep learning, essentially from scratch. The main takeaway and learning outcome will be that

Deep learning is a search for good hierarchical representations...

...that makes a given task easy to solve. The goal is to have everybody on board with this helpful description of deep learning and provide hands-on experience with how this translates into computer code (using PyTorch) via some concrete, simple examples.

👉 Get started here: Part 1: Building blocks.

In Part 2, we'll change gears and fly through more involved examples. The goal is to expose you to some of the many ideas, techniques, and tricks in modern deep learning. Those new to the field will hopefully get a useful impression of practical deep learning, with some pointers for learning more. Others more experienced in deep learning will perhaps learn new tricks and be exposed to exciting approaches and applications.

👉 Get started here: Part 2: Practical deep learning.

In Part 3, we look at how practical deep learning is part of software engineering, and what it takes to develop deep learning-based software solutions. This is an underappreciated but crucial aspect of deep learning and machine learning, more generally. Many components are necessary to actually deploy a machine learning-powered system and thus give it actual impact beyond a proof-of-concept research stage. In other words, machine learning engineering and deep learning engineering.

You can browse through the notebooks non-interactively using jupyter.org's nbviewer by clicking here.