Each week we'll be self-learning an important aspect of deep learning, with the suggested learning materials(not always provided) or your Googling. Then implement for your chosen project(can be one of the Kaggle competitions). In the weekly meetup, we'll go through participant's implementations and share/discuss important points.
The actual topic is subject to change:
- Week 1: overview
- Week 2: data preparation
- Week 3: CNN architecture
- Week 4: model training (loss functions + optimizer + training loop)
- Week 5: TOD
❗= Compulsory materials and assignments
- Learning goal:
- Acquire a broad overview of what deep learning is and the important steps (in the following week, we'll dive into each step in more detail and implement it yourself)
- Acquire an understanding of how to use fastai for quick deep learning implementation
- Mandatory material:
- ❗ Overview of Fastai: Fastai Lecture 1
- ❗ Overview of Deep learning: MIT: Introduction of Deeplearning
- Learn how to create virtual environments for your project
- conda env
- virtualenv
- Learn how to prepare data loaders in pytorch and fastai
- pytorch
- fastai
- Learn how to prepare your specific project data into a dataloader ready for deep learning models
You DON'T need to learn all these lectures before you start the kaggle project+meetup sessions. The idea is to learn ONE topic per week in depth and implement it yourself with Kaggle project.
You DON'T need to watch all of these courses, you may choose the course that appeals to you the most and use others as supplementary materials.
Previous meetup track on deep learning
Pick whichever project(s) that interests you.
- Reproduce a paper of your choice using the knowledge from the lectures
- Work on the following kaggle challenges